a systems analysis and technology roadmap for fall ...systemarchitect.mit.edu/docs/reddy20a.pdfmom,...

A Systems Analysis and Technology Roadmap for Fall Mitigation Systems for the Elderly

By

Vikas Reddy Enti Ranga Reddy Bachelor of Science in Mechanical Engineering

Cornell University, 2008

Submitted to the System Design and Management program in partial fulfillment of the requirements for the degree of

MASTER OF SCIENCE IN ENGINEERING AND MANAGEMENT

AT THE

MASSACHUSETTS INSTITUTE OF TECHNOLOGY

MAY 2020

© 2020 Vikas Reddy Enti Ranga Reddy. All Rights Reserved.

The author hereby grants to MIT permission to reproduce and to distribute publicly paper and electronic copies of this thesis document in whole or in part in any medium now known or hereafter created.

Signature of Author: __________________________________________________________________

Vikas Reddy Enti Ranga Reddy Fellow, System Design and Management Program

May 4th, 2020 Certified by: __________________________________________________________________

Bruce G. Cameron Director, System Architecture Group

Thesis Supervisor Accepted by: __________________________________________________________________

Joan S. Rubin Executive Director, System Design and Management

© 2020 Vikas Reddy Enti Ranga Reddy MIT SDM Thesis 2

[This Page Intentionally Left Blank]


A Systems Analysis and Technology Roadmap for Fall Mitigation Systems for the Elderly

By

Vikas Reddy Enti Ranga Reddy Submitted to the System Design and Management Program on May 4th 2020

in partial fulfillment of the requirements for the degree of Master of Science in Engineering and Management

Abstract Falls and fall related injuries in the elderly (aged 65 and older) are a major health challenge – both to the affected individual and to the public health system. Approximately 28-35% of the elderly fall each year and falls lead to 20-30% of mild to severe injuries, and are underlying cause of 10-15% of all emergency room (ER) visits. Falls cause 90% of the hip fractures in the elderly and also result in medical complications and high morbidity if the person does not receive prompt medical attention.

A fall mitigation system (FMS) is either a wearable or ambient system that detects falls, reduces fall related injuries and issues emergency alerts to prevent the long-lie. Current FMS have poor user adoption and are not as effective in preventing the long-lie. This thesis uses a systems approach to analyze architectures for a fall mitigation system architecture that can detect falls, reduce injury and issue emergency alerts to reliably prevent the long-lie in independent elders.

A National Health Interview Survey data was analyzed to understand the causes for falls, types of fall related injuries and common fall locations for community dwelling elders. A concept of operations was defined based on these findings and a user survey was conducted to understand the needs of community dwelling elders and the results were analyzed to prioritize system requirements for a fall mitigation system (FMS).

An FMS was decomposed into six level 2 functions and the various form choices for each of these functions were analyzed and rated for performance, power consumption and cost. Five different fall mitigation system architectures were analyzed and the Distributed-Hybrid architecture had the highest performance while the Integrated-Wearable architecture had the lowest power consumption.

Future technology trends in robotics, AI, neuromorphic computing and energy harvesting were studied to create a long-term strategic roadmap for fall mitigation systems. Neuromorphic architectures for computing and sensing offer the biggest performance per unit power unlock for fall mitigation systems.

Thesis Supervisor: Bruce G. Cameron

Title: Director, System Architecture Group & Lecturer System Design and Management


Acknowledgements

This thesis culminates a journey I started in Cambridge five years ago with several detours across London, Luxembourg, Boulder in between. I could not have managed this journey without the help of the people mentioned below.

Mom, Dad, Kutty - Thank you for keeping me grounded and for supporting me through all of life’s ups and downs.

My friends - Andres, Apoorva, Charles, Dev, Hock Boon, Leo, Nathan, Ryan, Steve and Yasho. Thank you for making life feel a little less busy and a lot more memorable.

Drago, I still have your email saved from 2015. I’ve done my best to follow #1 and I look forward to pursuing #2 on a much bigger scale.

“When you graduate from SDM, make sure to remember two things as a leader: (1) always stay true to yourself and (2) do good in/for this world.”

To my mentors, Amy, Parris and Scott - thank you for helping me navigate my life decisions. I appreciate your time, thoughtfulness and guidance. Thank you for keeping me focused on the long game.

L-team, M-team, Joe, Brad – I couldn’t have done this without your support and flexibility. Thank you for trusting me in my ability to deliver. To my teammates, thank you for your patience!

To my advisor, Bruce. I can’t thank you enough. You’ve been a tremendous guiding force throughout my MIT journey – right from your first Systems Architecture lecture to your most recent comment on my thesis.

And finally, to Caitlin. Thank you for being patient with me while I’ve been away in the attic. Your daily question, “What are you grateful for today?” has instilled a strong sense of gratitude for things I otherwise would have taken for granted. I’m grateful to have you in my life - you inspire me to be a better person.


Table of Contents ABSTRACT ....................................................................................................................................................... 3 ACKNOWLEDGEMENTS ................................................................................................................................ 4 TABLE OF CONTENTS .................................................................................................................................... 5 LIST OF FIGURES ........................................................................................................................................... 7 LIST OF TABLES ............................................................................................................................................. 9 1 INTRODUCTION .................................................................................................................................. 10

1.1 MOTIVATION ....................................................................................................................................... 10 1.2 FALL MITIGATION SYSTEMS - STATE OF THE ART .............................................................................. 12 1.3 CHALLENGES ................................................................................................................................... 14 1.4 PROBLEM STATEMENT AND SCOPE ................................................................................................... 15 1.5 THESIS OUTLINE .............................................................................................................................. 15

2 CONCEPT DEVELOPMENT OF A FALL MITIGATION SYSTEM ....................................................... 17 2.1 CONCEPT OF OPERATIONS V1 ............................................................................................................ 17 2.2 EPIDEMIOLOGY OF FALLS ................................................................................................................. 17

2.2.1 Dataset ..................................................................................................................................... 18 2.2.2 Methodology ............................................................................................................................ 18 2.2.3 Analysis ..................................................................................................................................... 19

2.3 CONCEPT OF OPERATIONS V2 .......................................................................................................... 25 3 SYSTEM REQUIREMENTS AND RECOMMENDATIONS ................................................................. 27

3.1 STAKEHOLDER ANALYSIS ................................................................................................................ 27 3.2 USER NEEDS SURVEY ...................................................................................................................... 28 3.3 HEALTHCARE FACTORS .................................................................................................................... 31 3.4 COST FACTORS ................................................................................................................................. 31 3.5 REGULATORY FACTORS .................................................................................................................... 31 3.6 REQUIREMENTS AND RECOMMENDATIONS ....................................................................................... 31

4 ANALYSIS OF SYSTEM ARCHITECTURE .......................................................................................... 33 4.1 FUNCTIONAL DECOMPOSITION ........................................................................................................ 33 4.2 EXISTING SYSTEMS ......................................................................................................................... 34

4.2.1 GoSafe2 .................................................................................................................................... 34 4.2.2 Hip’Safe .................................................................................................................................... 35 4.2.3 Kardian .................................................................................................................................... 36

4.3 ARCHITECTURAL DECISIONS ........................................................................................................... 37


4.4 ARCHITECTURE CONFIGURATIONS .................................................................................................. 39 4.4.1 Integrated Architecture ........................................................................................................... 39 4.4.2 Distributed Architecture .......................................................................................................... 41

4.5 ANALYSIS OF FUNCTIONS ................................................................................................................ 43 4.5.1 Powering .................................................................................................................................. 44 4.5.2 Sensing ..................................................................................................................................... 46 4.5.3 Computing ............................................................................................................................... 48 4.5.4 Communicating ....................................................................................................................... 50 4.5.5 Actuating .................................................................................................................................. 52 4.5.6 Interacting ............................................................................................................................... 55

4.6 TRADESPACE ANALYSIS ................................................................................................................... 56 5 TECHNOLOGY ROADMAP .................................................................................................................. 60

5.1 TECHNOLOGY TRENDS .................................................................................................................... 60 5.1.1 Neuromorphic Computing and Sensing .................................................................................. 61 5.1.2 Machine Learning Algorithms ................................................................................................ 63 5.1.3 Robotics .................................................................................................................................... 63 5.1.4 Energy Harvesting and Battery Improvements .................................................................... 65

5.2 ROADMAP ....................................................................................................................................... 67 6 CONCLUSION ....................................................................................................................................... 69 7 REFERENCES ........................................................................................................................................ 71 8 APPENDIX A: USER REQUIREMENTS SURVEY .............................................................................. 77 9 APPENDIX B: TRADESPACE ANALYSIS ............................................................................................ 79


List of Figures Figure 1 Global population growth projections of the Elderly (age >=65) [2] ................................................................. 10

Figure 2 Vicious cycle of falls in the elderly [9] ................................................................................................................. 11

Figure 3 Time on floor according to use of call alarm (143 falls in people aged >90 who were alone and unable to get up) [14] ............................................................................................................................................................................... 12

Figure 4 Overview of Fall Mitigation Systems. Purple boxes indicate high maturity of TRL7+ and Green boxes indicate medium maturity of TRL 3 to 6. ........................................................................................................................................ 13

Figure 5 Example of an ambient mmWave RADAR RF sensor for fall detection [19] .................................................... 14

Figure 6 V-Model of Systems Engineering Life Cycle [21] ............................................................................................... 15

Figure 7 CONOPS of a generic Fall Mitigation System ..................................................................................................... 17

Figure 8 Comparison of Age and Gender distribution between the US Census (2017) and the NHIS Survey ............... 20

Figure 9 Types of Injuries for people aged 65+ (source: NHIS dataset, 2015 to 2017) ................................................... 20

Figure 10 Cause of Fall (self-reported) for 65+ year olds (source: NHIS dataset, 2015 to 2017) .................................... 21

Figure 11 Intrinsic and Extrinsic factors for repeated Falls [28] ...................................................................................... 22

Figure 12 Injured body parts due to falls for 65+ year olds (source: NHIS dataset, 2015 to 2017) ................................ 23

Figure 13 Fall Related Injury types from 1998 to 2012 [29] ............................................................................................. 24

Figure 14 Location of falls for 65+ year olds (source: NHIS dataset, 2015 to 2017) ........................................................ 24

Figure 15 Location of Falls for 65+ year olds from the (Left) global WHO study [25] and (Right) Bergland study [30] ............................................................................................................................................................................................ 25

Figure 16 CONOPS v2 of a Fall Mitigation System ........................................................................................................... 26

Figure 17 Stakeholder Value Network for Fall Mitigation Systems for the Elderly ......................................................... 28

Figure 18 Result summary from the User Survey (Left to Right, Top to Bottom) 1) Age and Gender Distribution 2) Age at which respondents wanted a fall mitigation system 3) Fear of falling 4) Most-likely location for a fall (self-identified) 5) Fall mitigation system preference 6) Personal items that are always worn, even when sleeping ............................... 29

Figure 19 Functional decomposition of an FMS - Level 2. Fall Prevention is out of scope for this analysis. ................. 33

Figure 20 OPM of a generic Fall Mitigation System ......................................................................................................... 34

Figure 21 OPM of a PERS Fall Detection and Alerting System (boxes with red outlines indicate objects that are specific to this device) ..................................................................................................................................................................... 35

Figure 22 OPM of a Hip'Safe Fall Injury Reduction System (boxes with red outlines indicate objects that are specific to this device) ......................................................................................................................................................................... 36

Figure 23 OPM of a RF Fall Detection and Alerting System (boxes with red outlines indicate objects that are specific to this device) ......................................................................................................................................................................... 37

Figure 24 Integrated Architecture example for a Fall Mitigation System ........................................................................ 40

Figure 25 Distributed Modular Architecture for a Fall Mitigation System ...................................................................... 42

Figure 26 Power Consumption of Sensing and Communication in Wearable devices [63] ............................................ 47

Figure 27 Power consumption vs sampling rate for accelerometers and gyroscopes [64] .............................................. 47

Figure 28 Sensor mounting locations and fall detection accuracy [69] ........................................................................... 49


Figure 29 Data Rate vs Range of Wireless technologies [52] ........................................................................................... 51

Figure 30 Power consumption comparison of short-range wireless technologies [70] .................................................. 51

Figure 31 Power assist provided by an Exosuit [72] ......................................................................................................... 53

Figure 32 Wearable Fall Mitigation Systems (L to R) Controlled Moment Gyroscope [18], Hip'Safe Airbag [42], EPFL Exoskeleton [17] ................................................................................................................................................................. 54

Figure 33 A Walking Cane Robot as an Assistive Fall Mitigation System [16] ................................................................ 54

Figure 34 Smartphone Usage by Generation [73] ............................................................................................................ 56

Figure 35 Tradespace of Performance vs Power ............................................................................................................... 57

Figure 36 Tradespace of Performance vs Cost .................................................................................................................. 57

Figure 37 Dh1 – Distributed Hybrid Architecture OPM (boxes with red outlines indicate objects that are specific to this architecture) ....................................................................................................................................................................... 59

Figure 38 (left) Process types and performance [76] (right) Processor Performance vs Power consumption, 2019 [66] ............................................................................................................................................................................................ 61

Figure 39 Neuromorphic chips used as low-power event detectors [79] ......................................................................... 62

Figure 40 Fall Detection algorithm usage in literature (left) before 2014 (right) 2014 later [45] .................................. 63

Figure 41 Comparison of Object Detection Algorithms [82] ........................................................................................... 63

Figure 42 Comparison of Net Metabolic Change for Exoskeletons [83] .......................................................................... 64

Figure 43 Lim et al's Robotic Hip Exoskeleton [84] ......................................................................................................... 65

Figure 44 Energy density of Lithium battery chemistries [87] ........................................................................................ 66

Figure 45 Harvestable energy sources [89] ...................................................................................................................... 67

Figure 46 Roadmap of technology jumps for FMS ........................................................................................................... 68

Figure 47 FMS Performance/Unit Power Roadmap ......................................................................................................... 68


List of Tables Table 1 Description of the NHIS Falls dataset .................................................................................................................. 19

Table 2 Causes of falls in elderly adults: summary of 12 studies [26] .............................................................................. 22

Table 3 Risk Factors of Falls in the Elderly [26] ............................................................................................................... 22

Table 4: Summary of user needs for a Fall Mitigation system (1 - Least Important, 5 - Most Important) (n=78) ......... 30

Table 5 Categories of Requirements for Fall Mitigation Systems .................................................................................... 32

Table 6 System Requirements and Recommendations for a Fall Mitigation System ...................................................... 32

Table 7 Morphological Matrix of architecture decisions and form choices ..................................................................... 38

Table 8 Summary of Architectural Configurations ........................................................................................................... 39

Table 9 Options for Integrated-Wearable (Iw) Architecture ............................................................................................ 40

Table 10 Options for Integrated-Ambient (Ia) Architecture ............................................................................................ 41

Table 11 Options for Distributed-Wearable (Dw) architecture ........................................................................................ 42

Table 12 Options for Distributed-Ambient (Da) architecture .......................................................................................... 43

Table 13 Options for Distributed-Hybrid (Dh) architecture ............................................................................................. 43

Table 14 Design decisions and options for Powering ........................................................................................................ 44

Table 15 Rating of Power Storage and Recharging Systems ............................................................................................. 45

Table 16 Decisions and Options for Sensing ..................................................................................................................... 46

Table 17 Rating of Wearable and Ambient Sensing options ............................................................................................. 48

Table 18 Decisions and Options for Computing ............................................................................................................... 48

Table 19 Rating of Computing Systems ............................................................................................................................. 50

Table 20 Decisions and Options for Communicating ....................................................................................................... 50

Table 21 Rating of Communication Systems .................................................................................................................... 52

Table 22 Decisions and Options for Actuation/Injury Reduction .................................................................................... 52

Table 23 Rating of Injury Reduction Systems .................................................................................................................. 55

Table 24 Decisions and Options for Interacting/User Interface ...................................................................................... 55

Table 25 Rating of User Interface Systems ....................................................................................................................... 56

Table 26 Notable architectures for Fall Mitigation Systems ............................................................................................ 58

Table 27 List of impactful technologies for Fall Mitigation System Performance ........................................................... 60


1 Introduction

1.1 Motivation Falls and fall related injuries in the elderly (aged 65 and older) are a major health challenge – both to the affected individual and to the public health system. Approximately 28-35% of the elderly fall each year and falls lead to 20-30% of mild to severe injuries, and are underlying cause of 10-15% of all emergency room (ER) visits [1]. The major underlying causes for fall-related hospital admission are hip fracture, traumatic brain injuries and upper limb injuries [1].

Currently, there are ~800 million elders (9% of population) in the world and this number is projected to double to 1.6 billion (16% of population) by 2050 (Figure 1, [2]). Similarly, in the United States (US), the elderly population is projected to double from the current 52 million (16% of US population) to 95 million (23% of US population) in 2050 [3].

The cost of hospitalization for a fall related injury is ~$30k in the US [4]. In 2015, there were more than 3 million emergency room (ER) visits due to a fall and 28,000 fall-related deaths in the US. The estimated medical costs attributable to fatal and nonfatal falls was approximately $50.0 billion. For nonfatal falls, Medicare paid approximately $28.9 billion, Medicaid $8.7 billion, and private and other payers $12.0 billion [5]. With the projected growth of the elderly population, the healthcare costs for falls and fall related injuries are projected to increase to $240 billion in 2040. Consequently, the World Health Organization (WHO) has classified fall and fall related injuries as a major public health challenge that requires global attention [1].

Figure1GlobalpopulationgrowthprojectionsoftheElderly(age>=65)[2]


A fall is an event which results in a person coming to rest inadvertently on the ground or other lower level, not as a consequence of the following: sustaining a violent blow, sudden onset of paralysis, or an epileptic seizure [6]. Fall related injuries account of 90% of hip fractures in the elderly. Hip fractures are very concerning as they are associated with significant morbidity - the reported 1-year mortality after sustaining a hip fracture has been estimated to be 14% to 58% [7]. A recent meta-analysis revealed that women sustaining a hip fracture had a 5-fold increase and men almost an 8-fold increase in relative likelihood of death within the first 3 months as compared with age- and sex-matched controls [8].

Even if a fall does not result in an ER visit, it can often produce fear of falling resulting in a decrease in mobility, participation in activities, and independence. The long-lie i.e. involuntarily remaining on the ground for an hour or more following a fall often results in several medical complications such as dehydration, internal bleeding or even death. The long-lie was reported in up to 30% of falls and half of those who experience the long-lie die within 6 months of the fall [9]. Falls lead to this vicious cycle that affects the quality of life in the elderly and sometimes for their caregivers too. See Figure 2 for the causal factors and effects of falling in the elderly.

Figure2Viciouscycleoffallsintheelderly[9]

The elderly population (aged 65+) in the US will experience a growth bump of 55% in the next 15 years as the Baby Boomers (born between 1946-1964) age into 65+ category [5]. The Baby Boomers are tech-savvy, independent and value their identity. They are expected to age very differently compared to the Silent Generation (born between 1927-1945) with a strong preference for independent living and aging in place [10]. Elders who are aging in place or living independently in senior living communities are referred to as community dwelling elders. Elders who are in nursing homes are referred to as institution dwelling elders.

As aging in place becomes the norm with a growing population of community dwelling elders, it becomes even more important to address the causes of falls, reduce fall related injuries and to prevent the long-lie.


1.2 Fall Mitigation Systems - State of the Art Several methods are in use today to address the issue of falls and fall related injuries. There are three different functions involved in mitigating a person’s fall - Fall Detection (FD), Injury Reduction (IR) and Alerting. A Fall Mitigation System (FMS) incorporates a combination of functions from FD and IR to mitigate a person’s fall to prevent the long-lie. A Fall Prevention (FP) system prevents a person from falling and due to the multitude of underlying physiological conditions that cause a fall, fall prevention is not a generalizable solution [9]. This thesis will primarily focus on fall mitigation systems and not fall prevention.

There are several evidence-based methods to prevent falls including regular exercise, vitamin D supplementation and having regular fall risk assessments are actively deployed across nursing homes in the US [9]. Exercises that target balance, gait and muscle strength have been shown to reduce falls by 23% [11].

For community dwelling elders, the most common technology solution for fall mitigation is a Personal Emergency Response System (PERS) that is an on-demand call for help button that is either worn as a bracelet or as a pendant. There are an estimated 1.4 million users in the US [12]. However, the effectiveness of PERS is questionable as indicated by several studies [13][14]. Fleming et al investigated the usage of PERS in community dwelling elders and observed that despite the availability of a PERS pendant, the person who fell alone did not use their alarm to summon help 97% of the time. This resulted in a long-lie for 27% of the fallers and the distribution is shown in Figure 3. The leading cause for a person not using their alarm has been attributed to either a loss of consciousness or an inability to reach the button [14].

Figure3Timeonflooraccordingtouseofcallalarm(143fallsinpeopleaged>90whowerealoneandunabletogetup)

[14] Inthelastdecade,severalalternativestoPERShavebeeninvestigatedincludingawidearrayofwearableandambient

sensorsanddevicesforfalldetection,injuryreductionandfallprevention.

Figure 4 classifies the fall mitigation systems by their various functions and available forms along with their technology readiness levels (TRL). The Apple Watch Series 5 for example will automatically dial emergency services (911) if the watch detects a fall in a wearer older than 65.


Companies such as Kardian, Emerald, Walabot, CherryHome.ai and Care.ai are installing RF sensors and cameras in people’s homes to detect falls. Smart insoles from BioStride or Evone measure foot pressure differences to predict falls. Similar commercial progress is being made for injury reduction devices such as the hip protection airbags from Hip’Safe. The technology maturity for solutions in the fall prevention space is still very low for commercialization. The technologies in this space also range from vibrating insoles [15] to walking cane robots [16], exoskeletons [17] and backpacks with large control moment gyroscopes [18].

Figure4OverviewofFallMitigationSystems.PurpleboxesindicatehighmaturityofTRL7+andGreenboxesindicate

mediummaturityofTRL3to6.


Figure5ExampleofanambientmmWaveRADARRFsensorforfalldetection[19]

1.3 Challenges Even though there are several types of fall mitigation systems available for use today, their adoption is poor. Stokke et al analyzed the usability of PERS systems and found that the PERS devices were highly accepted but rarely used. Their analysis revealed that 25% of the respondents never wore the pendant. The reasons for non-compliance were comments such as: “forgot to put it on,” “worry it will get damaged,” “do not think they need it at the time,” “not satisfied with the PERS,” and “uncomfortable to wear”. They also found that all women interviewed who wore the pendant did so unwillingly. The PERS made it possible to live alone, but also made life more complicated due to choices as to when to wear and activate the alarm, and fear of triggering it by accident [13].

According to the Longevity Economic Outlook study commissioned by the AARP, Baby Boomers are less accepting of current fall mitigation systems due to their dated designs and the stigma associated with their age-specific usage [10]. Prof. Coughlin writes about this stigma and ageism in detail in his book, The Longevity Economy. Wearable devices such as the Apple Watch address this stigma by ‘hiding’ the fall mitigation system inside a multi-purpose, cool and age-neutral device. Non-wearable solutions such as Walabot’s RF sensors also mitigate this stigma by blending into home décor.

Wearable devices that work indoors and outdoors are preferred as they don’t constrain a person’s mobility. However, even accepted devices such as the Apple Watch face a fundamental challenge in ensuring that they are actually worn all the time. Since such devices require their batteries to be recharged every couple of days, users need to remove them periodically. There’s also a concern in the user group’s diligence in keeping the devices charged. Additionally, Stokke’s study shows that users were not comfortable in wearing the devices in their sleep or when using the bathroom. Hearing aid manufacturers such as Starkey are trying to address the usage challenge by embedding fall detection sensors into their advanced products with the hypothesis that users with hearing needs will always wear their hearing aids when they’re moving around [20]. No usage results are publicly available from Starkey to validate their hypothesis but again, Stokke’s study would dispute Starkey’s hypothesis.


Non-wearable solutions such as Walabot address the usage challenges by being always-on without requiring any user input. However, the assurance such devices provide is only valid inside a person’s home and as such, could limit a user’s mobility if they have had prior falls [9].

The above referenced studies indicate that for a fall mitigation system to be both accepted and used, it has to account for the changing needs of the user population. Initial analysis suggests that a fall mitigation system should be always-on, discreet and be able to monitor the user where ever they are. A user study in Chapter 3 will capture the evolving requirements in greater detail.

1.4 Problem Statement and Scope From the previous sections, the following points are clear:

1. Falls in the elderly is a major public health challenge and is a growing problem as Baby Boomers transition into old age.

2. Current solutions suffer from usability challenges and don’t prevent the long-lie. 3. Commercial proliferation of Internet of Things (IoT), Smart Homes and Robotics has

increased the availability of sensors, compute and actuators. However, these sensors and actuators have not made their way into commercially feasible fall mitigation systems.

This thesis will utilize a systems approach to analyze system requirements, understand system tradeoffs, conduct an analysis on alternative architectures and potentially make a recommendation for a fall mitigation system architecture that can detect falls, reduce injury and issue emergency alerts to reliably prevent the long-lie in independent elders. This thesis will then develop a technology roadmap by analyzing the technology trends in the IoT and Robotics industry and applying it to envision the future of fall mitigation systems.

1.5 Thesis Outline

Figure6V-ModelofSystemsEngineeringLifeCycle[21]

The V-model (Figure 6) is at the core of systems engineering and this thesis will traverse the left side of the V-model from concept development to analyzing the system architecture.

The thesis is organized as follows:


Chapter 2: Concept Development - This chapter will analyze falls data in greater detail to understand the primary causes and locations of falls in the elderly. This will inform the development of a concept of operations.

Chapter 3: System Requirements - This chapter will analyze stakeholder needs and discuss system requirements for fall mitigation systems based on user surveys and published literature.

Chapter 4: System Architecture – This chapter will decompose the functional architecture of various fall mitigation systems and compare them in context of the system requirements. A tradespace analysis will inform which architectures are performant.

Chapter 5: Technology Roadmap – This chapter will apply the technology roadmapping process and create a future looking roadmap for fall mitigation systems.

Chapter 6: Conclusion – The final chapter will summarize the key findings from the thesis and its limitations.


2 Concept Development of a Fall Mitigation System

The previous chapter highlighted the social and economic reasons for why falls have been classified as a major public health challenge. This chapter will analyze who is most susceptible to falls, why they fall, where they fall and the types of injuries from these falls. These findings will be used to develop a concept of operations (CONOPS).

2.1 Concept of Operations v1 A Fall Mitigation System (FMS) combines functions from fall detection and injury reduction systems into a form that is beneficial to the user. At a high level, an FMS is an automated system that applies the sense-process-act model to detect falls, trigger an injury reduction response and trigger an alert if the user is non-responsive. A solution-neutral concept of operations is shown in Figure 7. The key functions are: Fall Detection, Injury Reduction, User Interaction and Alerting.

Figure7CONOPSofagenericFallMitigationSystem

The design of the Fall Detection and Injury Reduction functions is highly dependent on the types of falls, location of falls and the body parts most likely to be injured. It is important to understand the epidemiology of falls to answer these questions in a statistically significant manner.

2.2 Epidemiology of Falls In this chapter, various datasets from the US Centers for Disease Control and Prevention (CDC) are analyzed in detail to understand the epidemiology of falls. Epidemiology is the study of how often diseases occur in different groups of people and why. Epidemiological information is used to plan and evaluate strategies to prevent illness and as a guide to the management of patients in whom disease has already developed [22].


2.2.1 Dataset

The National Center of Health Statistics (NCHS) is a CDC organization that conducts an extensive annual survey called the National Health Interview Survey (NHIS). The National Health Interview Survey (NHIS) is a household, face-to-face health survey of approximately 87,500 people in 35,000 households each year. The NHIS sample is designed to be representative of the civilian, non-institutionalized population living in the United States. By definition, the NHIS excludes residents in long-term care facilities, persons in correctional facilities, and U.S. nationals living abroad. Households comprised entirely of active-duty Armed Forces personnel are also excluded from the NHIS. Information is obtained about the health and other characteristics of each member of the household [23].

The NHIS contains many similar questions every year. The repeated items are called "core questions and are divided into three components -- Family, Sample Adult, and Sample Child. Supplemental questions and modules are added year over year. Questions about falls and injury due to falls have been included in the NHIS survey since 2004 and the data appears in the Injury and Poisoning Episode (InjPoE) dataset [23]. Verma et al used the InjPoE dataset in their study in 2016 to analyze the incidence rate of falls in the elderly and the injury rate[24]. This chapter uses the same dataset to identify the locations of falls, self-reported causes of falls and body parts most susceptible to fall injuries. 2.2.2 Methodology

Since the primary purpose of this analysis is to understand the frequency, location and injuries due to falls, it is important to analyze every single reported fall event. The InjPoE dataset contains all the required fields except for the age and gender of the respondent. These fields can be obtained by joining the InjPoE dataset with the Person dataset using the Household number (HHX), Family number (FMX) and Person number fields (FPX).

The analysis combines the Person and InjPoE datasets from 2015 to 2017 (2018 and 2019 datasets have not been released yet) to create a single dataset file that is easier to analyze and visualize. A python script was created to merge and recode the datasets into a single ‘nhis_falls.csv’ file. The script and the datasets have been uploaded to a publicly accessible Github repository, https://github.com/venti/FallMitigationProject. Table 1 describes the down selected fields in the nhis_falls.csv dataset which currently has 8877 records. This dataset was then further analyzed using Pandas/Python and Tableau and the key charts are shown in the next section.


Table1DescriptionoftheNHISFallsdataset

ID Variable Description Source DataType 1 SRVY_YR Survey Year NHIS Integer

2 HHX Household Number NHIS Integer

3 FMX Family Number NHIS Integer

4 FPX Person Number NHIS Integer

5 AGE_P Age NHIS Integer

6 gender Gender Recoded by author String

7 ICAUS Cause of Injury NHIS Integer

8 IJBODY1 1st Body part injured NHIS Integer

9 IJBODY2 2nd Body part injured NHIS Integer

10 IJBODY3 3rd Body part injured NHIS Integer

11 IJBODY4 4th Body part injured NHIS Integer

12 IFALL1 Location of 1st fall NHIS Integer

13 IFALL2 Location of 2nd fall NHIS Integer

14 IFALLWHY Cause of the fall NHIS Integer

15 injury_cause Cause of Injury Recoded by author String

16 body_part1 1st Body part injured Recoded by author String

17 body_part2 2nd Body part injured Recoded by author String

18 body_part3 3rd Body part injured Recoded by author String

19 body_part4 4th Body part injured Recoded by author String

20 fall_loc1 Location of 1st fall Recoded by author String

21 fall_loc2 Location of 2nd fall Recoded by author String

22 fall_reason Cause of the fall Recoded by author String

2.2.3 Analysis

2.2.3.1 Age Distribution

From 2015 to 2017, the NHIS has had a total of 279,090 respondents across the entire age range of 0 to 85+. The survey results have been binned by age groups in 5 year increments. For example, an age bin of 65 includes the age range of 65 to 69. From the US Census data (2017), 16% of the population is aged 65+ and the gender ratio is 55% female and 45% male for this age group. The NHIS data (2015 to 2017) has 16.1% of the population aged 65+ with a gender ratio of 55% female and 45% male for this age group. The age distribution of the survey respondents closely matches the age of the US population (US Census 2017 data) as shown in Figure 8 and can be treated as a representative dataset.

Of the 8877 injury records, 21% of the injury records (n=1834) are for the elderly (65+). The elderly have 30% more injuries per capita at 41 injuries per 1000 elderly respondents vs 31 injuries per 1000 survey respondent (all ages). Falls are self-reported to cause 64% of all injuries in the elderly and is the leading type of injury. The distribution of self-reported injury types can be seen in Figure 9. This finding is consistent with the WHO study where for ages 65 and older, the rate of fall injuries (serious enough to limit normal activities) was 47.7 per 1000 population [25].


Figure8ComparisonofAgeandGenderdistributionbetweentheUSCensus(2017)andtheNHISSurvey

Figure9TypesofInjuriesforpeopleaged65+(source:NHISdataset,2015to2017)

2.2.3.2 Fall Causes and Risk Factors

Figure 10 shows the self-reported cause of falls for the elderly (65+ years old). Slipping or tripping and losing balance or dizziness are the top two causes accounting for 75% of all falls in this dataset. While slipping and losing balance are the immediate causes for a fall, there are several underlying factors (intrinsic and extrinsic) that result in a person slipping or losing balance. Unfortunately, the NHIS dataset does not have information related to the underlying causes of a fall. The intrinsic and extrinsic factors for falls are shown in Figure 11.

Rubenstein et al [26] analyzed data from 12 studies that carefully evaluated elderly adults after a fall and assigned a ‘most likely’ cause and cataloged the results shown in

0.00%

0.50%

1.00%

1.50%

2.00%

2.50%

3.00%

3.50%

4.00%

0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85

Age and Gender Distribution: US Census 2017 vs NHIS Survey

Census-Female Survey-Female Census-Male Survey-Male


Table 2. A total of 3628 falls were analyzed across the 12 studies to calculate the mean percentage and the range shows the range of values across the 12 studies. The leading causes from Rubenstein’s study are ranked in the same order as the NHIS study (slipping/environment-related, losing balance, dizziness).

Figure10CauseofFall(self-reported)for65+yearolds(source:NHISdataset,2015to2017)

Falls are usually multifactorial in origin and since a single specific cause is hard to identify, it is recommended to identify risk factors instead [27]. Several prospective and retrospective epidemiology studies were analyzed by Rubenstein et al and based on 16 studies, weakness, balance deficit and gait deficit are the leading risk factors [26]. The rest of the risk factors are listed in Table 3. Based on the review of the risk factors and the various intrinsic factors for a fall, it is evident that a general-purpose fall mitigation system cannot effectively prevent falls. The thesis will continue to focus on fall detection, injury reduction and alerting to prevent the long-lie after a fall.


Table2Causesoffallsinelderlyadults:summaryof12studies[26]Cause Mean percentage (%) Range (%) ‘Accident’/environment-related 31 1–53 Gait/balance disorders or weakness 17 4–39 Dizziness/vertigo 13 0–30 Drop attack 9 0–52 Confusion 5 0–14 Postural hypotension 3 0–24 Visual disorder 2 0–5 Syncope 0.3 0–3 Other specified causes * 15 2–39 Unknown 5 0–21 * category includes arthritis, acute illness, drugs, alcohol, pain, epilepsy and falling from bed

Table3RiskFactorsofFallsintheElderly[26]Risk factor Mean Relative Risks Range Weakness 4.9 1.9–10.3 Balance deficit 3.2 1.6–5.4 Gait deficit 3 1.7–4.8 Visual deficit 2.8 1.1–7.4 Mobility limitation 2.5 1.0–5.3 Cognitive impairment 2.4 2.0–4.7 Impaired functional status 2 1.0–3.1 Postural hypotension 1.9 1.0–3.4

Figure11IntrinsicandExtrinsicfactorsforrepeatedFalls[28]


2.2.3.3 Fall Related Injuries

For effective injury reduction, it is important to understand the types of fall related injuries. The NHIS dataset has information about injured body parts from a fall and the distribution is shown in Figure 12. Injuries to the Head, Hip, Back, Knee and Shoulder account for 60% of the injured body parts and suggests that it is important for fall mitigation systems to protect these body parts. However, the NHIS dataset only captures the frequency of injured body parts and not the severity of injury.

Figure12Injuredbodypartsduetofallsfor65+yearolds(source:NHISdataset,2015to2017)

Min et al took a different approach to understand the severity of injuries and analyzed 2.8 million Medicare claims from 1998 to 2012 [29]. They were able to reliably classify 59000 records as fall related injuries (non-fatal) and the distribution is shown in Figure 13. Fractures to the legs, arms, trunk and open wounds to the head account for 70% of all fall related injuries. Min et al suggest that the fractures to the legs and arms are usually hip fractures and wrist fractures while the trunk fractures could either be to the shoulder or ribs [29]. Hip fractures are concerning as they are associated with significant morbidity - the reported 1-year mortality after sustaining a hip fracture has been estimated to be 14% to 58% [7]. A recent meta-analysis revealed that women sustaining a hip fracture had a 5-fold increase and men almost an 8-fold increase in relative likelihood of death within the first 3 months as compared with age- and sex-matched controls [8]. Thus, an effective fall mitigation system should definitely provide protection for the hip to prevent hip fractures from falls.


Figure13FallRelatedInjurytypesfrom1998to2012[29]

2.2.3.4 Fall Locations

The NHIS dataset was further analyzed to identify the leading locations for a fall. The choice of labels in the NHIS dataset is non-exclusive and since these locations are self-reported, it is possible for respondents to mis-classify the location of their fall. Falls on level ground and ‘Other’ account for 60% of the falls. The next cluster of falls happened on stairs (13%), transitioning on or off furniture (8%), on the sidewalk (7%) and in the bathroom (3%) and is shown in Figure 14.

Figure14Locationoffallsfor65+yearolds(source:NHISdataset,2015to2017)

A 2008 study by the WHO analyzed fall location data from a paramedic study of falls in Canada [25] and their results vary compared to the NHIS dataset. The results from the WHO study are more relevant for this thesis as the data is specifically related to falls that resulted in a call for an


ambulance and was captured immediately after a fall occurred. The WHO study concluded that the majority of falls (56%) occur outside the home. Inside the home, 26% of the falls happen on a level surface vs 6% on stairs and transitioning out of bed or a chair account for 3% of the falls each [25]. Another study conducted by Bergland captured the distribution of 300 elderly falls that all happened indoors [30]. It is unclear if these falls resulted in injuries or a call for emergency assistance. Figure 15 shows both the data from both the WHO study and the Bergland study.

Figure15LocationofFallsfor65+yearoldsfromthe(Left)globalWHOstudy[25]and(Right)Berglandstudy[30]

2.3 Concept of Operations v2 Based on the above analysis, for an FMS to be effective, it needs to be active when the elderly user is getting out of bed, using the shower, stairwells and also outside the home. An FMS needs to provide automatic alerting and also provide an activity log preceding the fall to help diagnose the cause of the fall. To reduce injuries, the FMS needs to protect the head, back, hips, shoulders and knees. Figure 16 shows the CONOPS of an integrated FMS that has both wearable and ambient equipment to detect falls and reduce injuries. The FMS communicates using wireless technologies to provide automatic alerts and to summon emergency services.

Outside56%Level Surface

26%

Shower or bath6%

Getting out of bed3%

On stairs6%

Chair3%

living rooms 30%

bedrooms29%

kitchens18%

bathrooms13%

hallways10%


Figure16CONOPSv2ofaFallMitigationSystem


3 System Requirements and Recommendations

This chapter will perform a stakeholder and needs analysis to help prioritize the requirements and recommendations for a Fall Mitigation System (FMS).

In the ISO/IEC Directives, a requirement is defined as an "expression in the content of a document conveying objectively verifiable criteria to be fulfilled and from which no deviation is permitted if compliance with the document is to be claimed." A recommendation is defined as an "expression in the content of a document conveying a suggested possible choice or course of action deemed to be particularly suitable without necessarily mentioning or excluding others" [31].

3.1 Stakeholder Analysis The requirements definition process begins with the identification of beneficiaries and stakeholders. A beneficiary benefits from the output of a product. A stakeholder is an individual or organization having a right, share, claim, or interest in a system or in its possession of characteristics that meet their needs and expectations [32][33].

An FMS has the user (typically aged 65+) as the primary beneficiary and several stakeholders as shown in the stakeholder value network in Figure 17. The user is the primary beneficiary for a fall mitigation system as he/she benefits from the effective functioning of the FMS. Additionally, the efficacy of the FMS depends on how well the system addresses the user’s needs.

The system boundary for the stakeholder value network is defined as first tier of stakeholders which includes the Family, the Healthcare providers (emergency services, doctors), the Emergency Response Control Center, Health insurance, Device manufacturer and the Government regulator. The Family is primarily interested in the health and wellbeing of the user. And depending on the age of the user, the family can sometimes be the decision maker and purchaser of the FMS [9]. The healthcare providers have professional interests in the efficacy of FMS to reduce the long-lie and to ensure better long-term health outcomes for the user. The Emergency Response Control Center expect timely alerts with zero false positives. The Health Insurance company will invest and pay for technologies and solutions that provide preventive care, improve health outcomes of the user and that reduce long-term healthcare costs. The Manufacturer is incentivized to produce an FMS that is reliable with high user satisfaction. The Government regulator cares about creating product regulations and guidelines that protect the user’s safety. In addition to the user’s needs, the needs of the stakeholders also need to be factored in when defining the requirements and recommendations of an FMS.

The minimum viable loop in the stakeholder value network that ensures a fallen user gets timely help and avoids the long-lie is highlighted in purple in Figure 17. This helps prioritize the beneficiaries and stakeholders where the User, as the primary beneficiary is the first priority followed by the device manufacturer and the emergency response control center. The healthcare provider was prioritized next as they influence the long-term health outcomes for the patient and can help address the physiological causes of the fall. The Family and Health Insurance are #4 as their primary role in reducing the long-lie is in providing financial assistance.


Figure17StakeholderValueNetworkforFallMitigationSystemsfortheElderly

3.2 User Needs Survey A survey was created to poll potential users to collect the various factors that they care about in a fall mitigation system. The questions in the survey were designed based on the learnings from Chapter 1 and Chapter 2 and has four sections. Section 1 collects non-identifiable information about the respondents age, gender, education level and location (country). Section 2 collects information about their usage of technology and artifacts they always wear, even in their sleep. Section 3 collects their responses to concerns about falling, location of falls and the type of fall monitoring system they would prefer. Section 4 collects their preferences for a fall mitigation system on a scale of 1 (least important) to 5 (most important) along the dimensions of privacy, style, multi-functionality, automatic alerting, on-demand alerting, mode of interaction, indoor and outdoor usage, injury reduction and if the system should be always-on.


Figure18ResultsummaryfromtheUserSurvey(LefttoRight,ToptoBottom)1)AgeandGenderDistribution2)Ageatwhichrespondentswantedafallmitigationsystem3)Fearoffalling4)Most-likelylocationforafall(self-identified)5)

Fallmitigationsystempreference6)Personalitemsthatarealwaysworn,evenwhensleeping

The survey was sent to respondents aged over 50 in the US, India and Singapore and a total of 79 responses were received. The survey was voluntary and anonymous and the raw data is available in the Github repository at: https://github.com/venti/FallMitigationProject.


A summary of the survey results is shown in Figure 18. About 85% (n= 67) of the respondents are aged between 55 and 74 and in this group there are 48% (n=32) female respondents. At least 46.8% (n=37) of the respondents indicated having a fear of falling. The stairs (42%) and bathroom (35%) were the most likely self-indicated locations for a fall and there were 6 null responses. This directionally agrees with the finding from the NHIS dataset analysis in Chapter 2.

There is a strong preference (76% of respondents, n=57) for a wearable fall mitigation system. The survey asked “Is there anything you always wear, even in your sleep?” and approximately 46% of the respondents (n=36) indicated underwear, followed by a watch/activity tracker (n=11, 14%). About 23% (n=18) respondents chose to not respond to that question. This suggests that a likely form for a fall mitigation system is to have it embedded in clothing articles such as underwear.

The results from Section 4 of the survey are shown in Table 4. The preference Score is computed by using Eq. 1 where n is the preference value (1 – least important and 5 – most important) and N(n) is the number of respondents for n.

cdefg = hij(i) Eq.1

The priority is assigned by sorting the Score in descending order with the highest score corresponding to the highest priority. A Fall Mitigation system designer can use these priorities to alter the system design for maximizing user satisfaction.

Table4:SummaryofuserneedsforaFallMitigationsystem(1-LeastImportant,5-MostImportant)(n=78)

Preferences N(1) N(2) N(3) N(4) N(5) Score Priority

Works both indoors and outdoors 0 0 7 16 55 360 1

Automatic Alerting 1 1 5 21 50 352 2

Always-on 0 0 6 26 45 347 3

On-demand alerting 2 5 8 22 41 329 4

Multi-Function 1 4 16 23 32 309 5

Interactive with buttons and lights 2 0 23 28 24 303 6

Interactive with a touchscreen display 3 10 15 24 25 289 7

Injury reduction 10 11 8 9 36 272 8

Style 5 7 25 20 19 269 9

Discreet 15 10 14 15 24 257 10

A few respondents reached out via email with further comments about their experience with current fall mitigation systems. A major complaint of current systems such as a wearable PERS with auto-alarming is that it will only trigger an alarm if it does not detect any movement after a fall. While this triggers on situations where a person might be unconscious after a fall, it fails to trigger the alarm for situations where a fallen person has chest movement but is still unable to manually trigger an alarm. A simple option here is to request the user to confirm they are OK after a fall and if no response is received within a timeout period, the device should trigger an alarm.


3.3 Healthcare Factors Non-fatal fall related injuries are a burden on the healthcare system with up to a 15% rehospitalization rate in the first month after being discharged [8]. The long-lie often results in several medical complications such as dehydration, internal bleeding or even death. The long-lie was reported in up to 30% of falls and half of those who experience the long-lie die within 6 months of the fall [9]. Falls cause 90% of the hip fractures which result in a 3-fold increase in 1-year mortality [7]. This suggests that injury reduction should be a much higher need for fall mitigation systems and will be used to prioritize requirement IR-1 in Table 6.

Medical insurance companies currently subsidize the cost of fall mitigation systems such as the PERS buttons and pendants [34]. If a proven fall mitigation system with injury reduction is available, it seems logical for insurance companies to cover those costs given.

3.4 Cost Factors There are two cost elements for Fall Mitigation systems – the cost of the device and the cost of the monitoring, alerting and communication service. PERS systems such as Philip’s GoSafe2 costs $99.95 for the device and $49.95/month for both indoor and outdoor fall monitoring [35]. Mounted systems such as Walabot also offer the device for free and charge $299/month for fall monitoring [19]. The Apple Watch does not have a monitoring service but requires an active cell phone service. The multi-functional device costs $500 and the monthly fees vary based on the cell phone provider.

Pricing is a business strategy decision and is out of scope of this thesis. It is however important to consider the accepted pricing of current devices and monitoring services when evaluating alternative architectures.

3.5 Regulatory Factors Regulations impact Device Manufacturers and affect their design and manufacturing choices. The topic is being mentioned in the thesis for completeness and is not meant to be a comprehensive analysis. In the US, Fall Mitigation systems are generally classified by the FDA as Class II medical devices and are regulated by Part 890, Physical Medical Devices of the Code of Federal Regulations Title 21 (21CFR). Each functional sub-system has to be evaluated and classified separately to comply with the stated regulations. For example, PERS devices are classified as either 21CFR890.3710 for a powered communication system and 21CFR890.3725 for powered environmental communication system and are exempt from pre-market notification [36], [37]. From an architecture standpoint, it is important to keep regulatory constraints in view and maintain simplicity and similarity where possible.

3.6 Requirements and Recommendations The results from the analysis of falls, analysis of user needs and stakeholder needs will be classified into the following categories: Fall Detection, Injury Reduction, User Interface, Alerting, Regulatory, Form Factor and System. They are defined in Table 5.


Table5CategoriesofRequirementsforFallMitigationSystems

Category Code Definition Fall Detection FD Requirements related to the fall detection function Injury Reduction IR Requirements related to the injury reduction function User Interface UI Requirements related to the user interface Alerting ALR Requirements related to the alerting mechanisms Regulatory REG Requirements related to regulatory compliance Form Factor FF Requirements related to the form factor System SYS Requirements that apply across the fall mitigation system

The following requirements are generated based on literature review in Chapter 1, and the analysis of epidemiology of falls (Chapter 2), concept of operations (Chapter 2) and the user needs survey (Chapter 3).

Table6SystemRequirementsandRecommendationsforaFallMitigationSystem

ID Name Description

ALR-1 Automatic Alerting The FM system shall automatically trigger an alert if it detects a fall and does not receive a response from the user within <X> seconds

ALR-2 On-demand alerting The FM system shall trigger an alert based on input from the user

FD-1 Environment The FM system shall detect falls in under 100ms both in indoor and outdoor environments with less than 0.1% false positives [38]

FD-2 Always-on The FM system shall detect falls during all activities of daily life including showers

FD-3 Accuracy The FM system shall have a fall detection accuracy of at least 99% with less than 0.001% false negatives [39]

FF-1 Wearable The FM system should be wearable by the user

FF-3 Discreet The FM system shall be worn or installed in a discreet manner so as to not draw attention to itself from non-users.

IR-1 Injury reduction The FM system should reduce fall-related injuries for the user's head, neck, shoulder, hips or knees

IR-2 Reuse The FM system should be reusable with at-home changes after a fall

REG-1 FDA The FM system shall comply with 21CFR890.3710 and 21CFR890.3725

SYS-1 Cost The FM system should cost less than $1000 and the monthly service fee shall be less than $100/month

SYS-2 Reliability The FM system shall be reliable with a mean time between failure (MTBF) of 35000 hours (4 years)

SYS-3 Multi-Function The FM system should be expandable to provide additional health monitoring functions

SYS-4 Interoperability The FM system should be inter-operable with smartphones, smart home systems and other assistive devices

SYS-5 Battery Life The FM system shall be rechargeable when in use

UI-1 Interactive The FM system should be interactive by the user or caregiver with either direct tactile control or remote control

The next step is to map these requirements to the various functions of a Fall Mitigation System and analyze the tradespace of systems architectures.


4 Analysis of System Architecture

This chapter will evaluate the architectures of existing fall mitigation systems, analyze their functions and will generate a tradespace of new architectures that can address the system requirements and user needs identified in Chapter 3.

Architecture is defined as the embodiment of concept, and the allocation of physical/informational function to elements of form, and definition of relationships among the elements and with the surrounding context [33]. The first step in the analysis of system architecture is to decompose the system’s functions and form.

4.1 Functional Decomposition The CONOPS in Chapter 2 identified four Level 1 functions for an FMS: Fall Detection, Injury Reduction, Alerting and User Interface. The Level 1 functions have been further decomposed to six Level 2 functions - Sensing, Computing, Actuating, Interacting, Communicating and Powering and is shown in Figure 19. An Object Process Model (OPM) for a generic Fall Mitigation system was created and is shown in Figure 20.

Figure19FunctionaldecompositionofanFMS-Level2.FallPreventionisoutofscopeforthisanalysis.


Figure20OPMofagenericFallMitigationSystem

4.2 Existing Systems A commercially available system that reliably performs all three core functions of fall detection, reducing injury and emergency alerting to prevent the long-lie in elders does not exist today. Three systems come close to addressing a subset of these functions and their architectures will be analyzed to identify form choices and infer architectural decisions. 4.2.1 GoSafe2

An advanced PERS system such as Philips’s GoSafe2 is a single-purpose system that uses accelerometers, barometric sensors and tuned fall detection algorithms to reliably detect falls and automatically issue emergency alerts via an inbuilt cellular radio [40]. The device also has onboard GPS sensors and provides the device’s location as part of the alert to enable emergency responders to find the fallen user easily. The GoSafe2’s product webpage lists several customer reviews and a common complaint is that the pendant device is heavy and bulky to be hanging from someone’s neck for the whole day. Additionally, reviewers complain about user compliance as there is a cognitive load involved in remembering to recharge the device (every 2 to 3 days) and to also wear it again after a person removes it when sleeping [35]. These findings are consistent with the discussion in Chapter 1 – a PERS system alone cannot reliably prevent the long-lie.


Figure21OPMofaPERSFallDetectionandAlertingSystem(boxeswithredoutlinesindicateobjectsthatarespecificto

thisdevice)

4.2.2 Hip’Safe

Passive injury reduction systems such as foam pads woven into underwear have proven to reduce impact forces from falls but their user acceptance and compliance is low due to issues with comfort and stigma associated with the ‘bulky-hip’ appearance [41]. The Hip’Safe airbag system from Helite is currently the only commercially available active injury reduction system that both detects a fall and also deploys an airbag to reduce injury to the hips [42]. The Hip’Safe uses accelerometers, gyroscopes and fall detection algorithms to detect falls in under 200ms and triggers the airbag actuator to inflate the airbag in under 80ms both indoors and outdoors. The inflated airbag reduces impact forces to the hips from the fall by ~90% [42] and has been verified in an independent test [43]. The device is bulky and weighs 1kg and has a rechargeable battery that lasts 7 days. A major shortcoming of the Hip’Safe device is that it cannot be used in the shower and its bulkiness prevents long-term daily use. The cognitive load for recharging and wearing the device is similar to a PERS system. Additionally, the device does not have an alerting feature and thus it cannot be solely used to prevent the long-lie. Startups such as TangoBelt have recently introduced a similar wearable airbag solution but details are scarce [44].


Figure22OPMofaHip'SafeFallInjuryReductionSystem(boxeswithredoutlinesindicateobjectsthatarespecifictothis

device)

4.2.3 Kardian

There are a class of radio-frequency (RF) based systems (Walabot, Emerald, Kardian) that use different frequency bands (RADAR, Ultra-Wide-Band, Wifi) to detect falls and monitor a person’s vitals. The Kardian device is a discreet ambient sensor, has a similar footprint of a smoke detector and can be installed in all rooms of a home. The device constantly monitors people’s pose and can detect abrupt and slow sliding falls by analyzing the returned RF signals. In parallel, the device is also remotely monitoring heart rate. By combining the pose and body vitals information, the Kardian device can detect falls and issue automatic alerts with very few false positives compared to wearable systems [19]. These systems have no cognitive load as they are always on with wall-power. They can be used to reliably prevent in the long-lie inside people’s homes as long the installation covers all rooms effectively.


Figure23OPMofaRFFallDetectionandAlertingSystem(boxeswithredoutlinesindicateobjectsthatarespecifictothis

device)

4.3 Architectural Decisions The following morphological matrix has been generated based on the analysis of form choices and architectural decisions from existing systems in the previous section and from the systems listed in the state-of-the-art section in Chapter 1. The decisions are numbered D1 to D19 and the options range from A1 to A5 as shown in Table 7. The global architecture decisions affect the overall configuration of the system and are discussed in detail in the next section on Architecture Configurations. One limitation to the analysis here is that each of the options are designed to be mutually exclusive to simplify the tradespace analysis. Each of the options is discussed in detail and is assigned a relative cost, power and performance rating in the Analysis of Functions section.


Table7MorphologicalMatrixofarchitecturedecisionsandformchoices

ID Scope Architectural Decision Alt. 1 (A1) Alt. 2 (A2) Alt. 3 (A3)

Alt. 4

(A4)

Alt. 5

(A5)

D1 Global Sensing Mode Wearable Ambient Hybrid Sens

D2 Global Injury Reduction Mode Wearable Inj Non-Wearable None

D3 Global Computing Mode Onboard Offboard Hybrid Comp None

D4 Global Number of devices Single Multiple

D5 Global Communication Mode Short-Range Long-Range Hybrid Comms None

D6 Global Alerting Local Control Center Hybrid Alert None

D7 Sensing Wearable Sensing

Accelerometer

Gyroscope + Acc.

Barometer + Acc. None

D8 Sensing (Non-Wearable) Ambient Sensing

mmWave RADAR UWB Camera IR None

D9 Powering Storage

Lithium Battery

Supercapacitor None

D10 Powering Recharging TEG Wireless Charging Wall Kinetic None

D11 Communicating Short-Range BLE ANT+ Zigbee Wifi None

D12 Communicating Long-Range LTE-M LoRA SigFox 4G None

D13 Computing Onboard Tensor Chip Mobile chip FPGA ASIC None

D14 Computing Offboard Cloud Edge None

D15 Computing Algorithms Threshold ML - kNN/SVM Deep Learning

D16 Interacting Input Buttons Touchscreen Mic Smartphone None

D17 Interacting Output LEDs Display Speaker Smartphone None

D18 Actuating Wearable Actuation Airbag Exoskeleton Exosuit CMG None

D19 Actuating Non-Wearable Actuation

Robotic Cane Foam Pads

Mechanical Walker None


4.4 Architecture Configurations Based on the architectural review of the three existing systems in Section 4.2 and based on the review of several comparative studies of fall detection systems ([45]–[48]), there are two distinct approaches to fall mitigation systems’ architecture: Integrated and Distributed and with two distinct forms: Wearable and Ambient.

Table8SummaryofArchitecturalConfigurations

Integrated (I) Distributed (D)

Wearable (w) Iw

PERS (GoSafe2)

Hip’Safe

Dw

Shimmer Motes [49] (<=TRL6)

Ambient (a) Ia

Robotic Cane [16] (<=TRL6)

Mechanical Walker

Da

Kardian

Care.AI, Walabot, Emerald [50]

Hybrid (h) Dh

Smart Home Integration [51] (<=TRL6)

4.4.1 Integrated Architecture

An integrated architecture tightly couples all the six FMS functions on a single device. This is the current architecture of wearable systems such as PERS and Hip’Safe where there is a non-modular, single purpose system that requires time-sensitive coordination between its various sub-systems to deliver fall detection, injury reduction and alerting functions. An integrated architecture is simpler, has fewer components and can be compact. However, it lacks flexibility, and extensibility. It is not feasible to extend the architecture to create a multipurpose device without re-engineering the entire device.


Figure24IntegratedArchitectureexampleforaFallMitigationSystem

The two types of Integrated architecture for fall mitigation systems are Integrated-Wearable (Iw) and Integrated-Ambient (Ia) and their relevant decision options are highlighted in green in Table 9 and Table 10. Note that in both architectures, the computing mode has to be onboard and there can only be a single device.

Table9OptionsforIntegrated-Wearable(Iw)Architecture


Alt. 4 (A4)

Alt. 5 (A5)


D2 Global Injury Reduction Mode

Wearable Inj

Non-Wearable None



D5 Global Communication Mode Short-Range Long-Range

Hybrid Comms None



Table10OptionsforIntegrated-Ambient(Ia)Architecture


Alt. 4 (A4)

Alt. 5 (A5)



Wearable Inj

Non-Wearable None




Hybrid Comms None


4.4.2 Distributed Architecture

A distributed architecture creates a new M2M (machine to machine) interface function that is embedded inside the existing communicating function and enables the distribution of the six functions across several modules that can either be packaged into a single device or split across multiple devices. The M2M interface can connect and network the modules either over a physical connection or over a wireless BLE network.

This architecture has the benefit of modularity and enables the addition of new functions easily but incurs the complexity of maintaining commonality across the devices. The RF sensing systems have this architecture where several RF sensors can be installed across a home but they still function as a single system. This architecture enables the creation of a hybrid fall mitigation system consisting of multiple wearable devices and remote monitoring devices that work in unison to reduce false positives and fill the monitoring gaps in areas such as the shower and when devices are not being worn indoors [51]. New functions can be added easily by interfacing new hardware modules such as a heart rate monitoring module or by adding new software via smartphone apps connected via BLE to the system. In this architecture, injury reduction for several body areas can be achieved by creating several injury reduction modules that connect to the user’s BLE network and are mounted close to the body areas of interest.


Figure25DistributedModularArchitectureforaFallMitigationSystem

The three types of Distributed architectures are Distributed-Wearable (Dw), Distributed-Ambient (Da) and Distributed-Hybrid (Dh) and their available architectural configurations are highlighted in green in Table 11, Table 12 and Table 13.

Table11OptionsforDistributed-Wearable(Dw)architecture


Alt. 4 (A4)

Alt. 5 (A5)



Wearable Inj

Non-Wearable None




Hybrid Comms None



Table12OptionsforDistributed-Ambient(Da)architecture


Alt. 4 (A4)

Alt. 5 (A5)



Wearable Inj

Non-Wearable None




Hybrid Comms None


Table13OptionsforDistributed-Hybrid(Dh)architecture


Alt. 4 (A4)

Alt. 5 (A5)



Wearable Inj

Non-Wearable None




Hybrid Comms None


4.5 Analysis of Functions The six functions and associated decisions (D7-D19) will be discussed in detail below. To simplify the tradespace analysis, the user preferences and system requirements have been synthesized down to three dimensionless metrics:

• Cost • Power Consumption • Performance


The various options are then assigned relative scores on the modified Fibonacci scale (0,1,2,3,5) for each of the metrics. The modified Fibonacci scale was used to indicate to both capture a null score (0) and to also capture a significant change in a metric, especially when it went from 3 to 5. The Cost metric (1 means lower cost, 5 means greatest cost) includes the bill of material cost, development cost, integration cost and maintenance cost. The Power Consumption metric (1 means lower power consumption, 5 means highest power consumption) captures the relative power consumption. The Performance metric (1 means lower performance, 5 means highest performance) captures the relative performance of an option in terms of the author’s assessment of the option’s impact on Fall Detection accuracy, Injury Reduction Efficacy and User Compliance based on literature research. For options that lack data from literature, the author has made a conservative guess.

4.5.1 Powering

The powering function captures all aspects of energy storage and recharging for an FMS and has two decisions associated with it (Table 14). Decision D9 deals with the energy storage choice and decision D10 deals with the recharging or direct powering options.

Table14DesigndecisionsandoptionsforPowering

ID Scope Architectural Decision Alt. 1 (A1) Alt. 2 (A2)

Alt. 3 (A3)

Alt. 4 (A4)

Alt. 5 (A5)

D9 Powering Storage Lithium Battery Supercapacitor None

D10 Powering Recharging TEG Wireless Charging Wall Kinetic None

For an ambient FMS system, even if the primary power source is from wall-power, it is preferred to have an internal battery for redundant power backup to ensure high system availability (FD-2 requirement). For wearable systems, power consumption is the primary system level constraint as it limits the usage time for a device and this constraint is mainly driven by the limitations in energy density of batteries.

For energy storage, batteries and supercapacitors are the two common choices for sensor systems [52]. It is desirable to have a battery that is both small and light with the highest energy density. The energy density of several batteries is shown in Figure 44. The ideal technology is on the top right corner of the chart with the highest volumetric and mass energy density. Rechargeable Lithium batteries are currently the preferred choice for electronic devices due to their high cell voltage (3.6+V) and high energy density compared to Lead acid, NiMH and NiCD solutions [53].

Supercapacitors are commonly used wherever a quick energy boost is needed, as an alternative to a rechargeable battery. As their energy density increases, supercapacitors are becoming the preferred energy storage system for energy harvesting devices, as they are quickly rechargeable, lightweight and don’t self-discharge [54]. They can store the energy generated and produce it quickly when, say, wireless transmission is needed.

From the CONOPS in Chapter 2, the typical operations lifecycle for wearable systems involves a time period where the device cannot be used as it is being recharged. For an FMS, this poses a cognitive burden on the elderly user to remember to recharge the device and to remember to wear


it once the device is charged. Besides the traditional approach of connecting a device to wall power, other options for recharging include energy harvesting with either wireless charging over long range (RF or IR [55]), thermoelectric generators or kinetic energy recovery systems. Recharging systems need to provide between 50mWh to 100mWh of energy to recharge batteries and supercapacitors and it can happen in short bursts or over prolonged periods of time as a trickle charger.

Wireless charging technologies from companies such as Wi-Charge (IR based) and Witricity (RF based) are able to remotely power and charge low-power systems at distances over 5m. Such a system can enable recharging of an FMS while the device is still in use when worn by the user, for example during their sleep. Wireless chargers can deliver between 100mW to 1W of power at distances of 1m to 5m [56].

Thermoelectric Generators (TEG) are a class of energy harvesting systems that use temperature differentials across a thermocouple to generate a voltage and power a device. Companies such as Matrix Industries have patented thermoelectric generators than can power wearable devices by harvesting the temperature difference between a user’s skin and the environment [57]. These TEGs generate 30 µW/cm2 from human skin contact as long as the standing air temperature is below 32°C and a TEG the size of a watch can generate ~0.8mW of power which can be used as a continuous trickle charger [58].

Kinetic recovery systems convert kinetic energy into electrical energy. They are typically used in wrist watches and can harvest between 10mW to 250mW. These systems are preferred for wearable systems worn at the extremities (wrist, ankle) and are not as effective in generating power when worn on the torso [59].

Based on the above analysis, ratings have been assigned to the storage and recharging options and are shown in Table 15.

Table15RatingofPowerStorageandRechargingSystems

Function Type Options Cost Power Consumption FMS Performance

Powering Storage Lithium Battery 2 0 5

Powering Storage Supercapacitor 1 0 2

Powering Storage None 0 0 0

Powering Recharging TEG 5 0 1

Powering Recharging Wireless Charging 3 0 3

Powering Recharging Wall 1 0 5

Powering Recharging Kinetic 3 0 2

Powering Recharging None 0 0 0


4.5.2 Sensing

Sensing is the function of detecting events or changes in the subject or environment and sending the information to a computer processor [60]. There are three architectural decisions related to sensing: D1 for sensing mode, D7 for wearable sensing options and D8 for ambient sensing options (Table 16). Fall sensing today is primarily achieved with inertial sensing and barometric sensing for wearable systems [61], [62]. For ambient systems, RF sensing and vision sensing are the common options.

Table16DecisionsandOptionsforSensing


Alt. 4 (A4)

Alt. 5 (A5)


D7 Sensing Wearable Sensing Accelerometer

Gyroscope + Acc.

Barometer + Acc. None

D8 Sensing Ambient Sensing

mmWave RADAR UWB Camera IR None

A common sensor configuration for wearable fall mitigation systems is to use a triaxial accelerometer running at a sampling rate of 10 Hz to 100 Hz ([46], [46], [47]). Sensing accounts for 50% of the power consumption in a wearable sensor device (Figure 26, [63]) and it is critical to choose sensing options that are power efficient. The power consumption of accelerometers increases with the accelerometer’s accuracy and the sampling rate as shown in Figure 27. A high accuracy accelerometer has between a 10x to 100x increase in power consumption compared to a low accuracy accelerometer. Since the power budgets are a defining constraint for wearable systems, one strategy for power-efficient inertial sensing is to pair both a high accuracy and low accuracy accelerometer and to use event triggers from the low accuracy accelerometer to power on the high accuracy accelerometer to take measurements [64]. Other alternatives include coupling a barometric sensor with a low power accelerometer to provide high accuracy fall sensing [61]. Gyroscopes use atleast 100x more power than accelerometers but are beneficial for fall sensing where it is important to know the direction of the fall, typically in systems that also have injury reduction mechanisms. In such scenarios, the gyroscope is coupled with an accelerometer and has both a high cost and high power consumption for slightly better performance.


Figure26PowerConsumptionofSensingandCommunicationinWearabledevices[63]

Figure27Powerconsumptionvssamplingrateforaccelerometersandgyroscopes[64]

The RF sensors track the fluctuation of radio frequency signals (RADAR, UltraWideBand) or wireless channel state information (such as WiFi, Bluetooth) to detect or predict falls, as the strenuous body movement speed brings out abnormal changes on RF signals. mmWave RADAR systems have more readily available chipsets due to their usage in the automotive industry for obstacle detection and have lower power consumption and high detection performance [46]. UWB chipsets have historically been used for location and are now being repurposed for human tracking and fall detection. The performance of UWB systems is lower compared to mmWave RADAR and the systems also consume more power [48].

Vision based sensing options include cameras and infrared imaging (IR). Cameras are the cheapest sensors with low power consumption. However, there is a high computational cost with low performance compared to RF systems. IR sensing is the most expensive option and it also has high power consumption with relatively low fall detection performance compared to cameras and RF sensors [46].

Based on this analysis, the ratings for wearable and ambient sensing are assigned as shown in Table 17.


Table17RatingofWearableandAmbientSensingoptions


Sensing Wearable Accelerometer 1 1 2

Sensing Wearable Gyroscope + Acc. 3 3 3

Sensing Wearable Barometer + Acc. 5 5 5

Sensing Wearable None 0 0 0

Sensing Ambient mmWave RADAR 3 2 5

Sensing Ambient UWB 3 3 3

Sensing Ambient Camera 1 1 3

Sensing Ambient IR 5 3 2

Sensing Ambient None 0 0 0

4.5.3 Computing

Computing is defined as the function that uses computer processors and algorithms to manage, calculate and communicate information. In an FMS, the computing function is primarily responsible for processing raw sensor data to identify fall events and trigger programmed actions such as activating injury reduction mechanisms and emergency alerting mechanisms. The key decision for computing is D3 on whether to perform all computing activities onboard, offboard or have a hybrid model. Decisions D13 and D14 cover the options of computing processor types and decision D15 covers the algorithm options.

Table18DecisionsandOptionsforComputing


Alt. 4 (A4)

Alt. 5 (A5)


D13 Computing Onboard Tensor Chip Mobile chip FPGA ASIC None

D14 Computing Offboard Cloud Edge None

D15 Computing Algorithms Threshold ML - kNN/SVM

Deep Learning None

The computing function typically comprises of a microprocessor, memory, storage and algorithms. The microprocessor’s performance and power consumption are important parameters for an FMS as it affects the requirement FD-1 for detecting a fall in under 100ms.

A hybrid computing option is a common option for decision D3 for wearable systems. The wearable device is used as a data logger and cloud servers are leveraged to process the data to


detect and classify events after the fact [65]. While this scheme works for applications that require historical analysis such as fitness trackers, the latency involved in the round trip data transfer and cloud processing is in the order of seconds and far exceeds the 100ms detection requirement in FD-1. The fall detection computation needs to happen onboard but other functions such as classification, training, user calibration can happen offboard. Tensor chips are designed for AI/ML algorithms and have high performance at a high power consumption. Mobile chips are general purpose processors that can handle linear computation functions efficiently. FPGA (Field Programmable Gate Arrays) chips implement logic in hardware and have a very high performance for low power consumption but have a high development cost. Finally, ASICs (Application Specific Integrated Circuits) offer very high performance for lower power consumption but have high development costs with minimal flexibility [66].

For Offboard computing, the Cloud option refers to high performance servers in remote data centers while the Edge option refers to on-premise embedded servers. Edge computing has better power efficiency and higher performance due to lower latency compared to cloud computing but comes at a higher system cost [67].

The performance of Fall Detection algorithms is just as important to ensure that the FMS can meet the FD-1 requirement. The two common types of algorithms are threshold based or machine learning (ML) based. Threshold based algorithms compare sensor values against previously configured thresholds to determine if an event is a fall. Threshold based methods are power efficient but have a high degree of false positives. ML based algorithms primarily use support vector machines (SVM) or k-nearest neighbors (k-NN) and have both higher power consumption and higher accuracy compared to threshold based methods [68]. A study by Ozdemir et al concluded that the best single-sensor performance for fall detection (99.96% sensitivity, 99.87% accuracy and 99.76% specificity) is achieved with a waist mounted sensor using the k-NN algorithm [69]. For the tradespace analysis, ML algorithms are ranked higher for performance compared to Threshold algorithms. Deep learning is a specific class of machine learning algorithms that uses neural networks to accurately classify non-time series, 2D and 3D data such as images and point clouds generated by RF and vision sensors. It has a high power consumption but also high performance for such datasets [39].

Figure28Sensormountinglocationsandfalldetectionaccuracy[69]

Based on the above analysis, the ratings for the various options for decisions D13, D14 and D15 are listed in Table 19.


Table19RatingofComputingSystems


Computing Onboard Tensor Chip 5 5 5

Computing Onboard Mobile chip 1 3 2

Computing Onboard FPGA 3 3 5

Computing Onboard ASIC 3 1 5

Computing Onboard None 0 0 0

Computing Offboard Cloud 3 5 3

Computing Offboard Edge 5 3 5

Computing Offboard None 0 0 0

Computing Algorithms Threshold 1 1 3

Computing Algorithms ML - kNN/SVM 3 3 5

Computing Algorithms Deep Learning 5 5 5

Computing Algorithms None 0 0 0

4.5.4 Communicating

The communicating function for FMS covers both short and long range inter-device communication via wireless means. The key decisions are D5, D11 and D12 but the decisions made in D3, D4 and D6 also affect the communicating options. If D3 is set to either offboard or hybrid, or if D4 is set to multiple devices, or if alerting is set to control center or hybrid, then a communicating option is required in D11 or D12.

Table20DecisionsandOptionsforCommunicating


Alt. 4 (A4)

Alt. 5 (A5)



D5 Global Communication Mode

Short-Range Long-Range

Hybrid Comms None


D11

Communicating Short-Range BLE ANT+ Zigbee Wifi None

D12

Communicating Long-Range LTE-M LoRA SigFox 4G None


Figure29DataRatevsRangeofWirelesstechnologies[52]

The relevant wireless technologies and their data rate vs range is shown in Figure 29. ANT, ZigBee and Bluetooth Low Energy (BLE) are common solutions for wireless sensor networks operating in the 10 mA to 30mA peak current consumption for a communication range of up to 30m. WiFi is more efficient if a constant high data throughput is required but it consumes 10x more current and is not the best choice for wireless sensor networks [70]. BLE has the best power vs performance at 0.153 uW/bit when controlled for peak current consumption of under 50mA (Figure 30).

For long-range communication, the growth in IoT (internet of things) sensors has prompted the creation of several industry standards for low-power long range radios. The leading technologies in 2020 are LTE-M, Sigfox and LoRa. For a wearable system, LTE-M has the best performance in terms of bandwidth, latency and power consumption [52].

Figure30Powerconsumptioncomparisonofshort-rangewirelesstechnologies[70]

The ratings for the various options are applied based on the above analysis and is shown in Table 21.


Table21RatingofCommunicationSystems


Communicating Short-Range BLE 3 1 3

Communicating Short-Range ANT+ 1 2 1

Communicating Short-Range Zigbee 1 2 2

Communicating Short-Range Wifi 3 5 5

Communicating Short-Range None 0 0 0

Communicating Long-Range LTE-M 3 1 5

Communicating Long-Range LoRA 2 1 2

Communicating Long-Range SigFox 2 1 2

Communicating Long-Range 4G 5 5 3

Communicating Long-Range None 0 0 0

4.5.5 Actuating

The actuating function is required to activate injury reduction mechanisms that are either wearable or non-wearable. Wearable injury reduction mechanisms include inflatable airbags, exoskeletons, exosuits or control moment gyroscopes (CMG). Wearable foam pads have not been considered due to the stigma associated with their use and low user compliance [41]. Non-wearable injury reduction mechanisms include foam pads for the floors, mechanical walkers and robotic canes. The architectural decisions and the options are shown in Table 22.

Table22DecisionsandOptionsforActuation/InjuryReduction


Alt. 4 (A4)

Alt. 5 (A5)


Wearable Inj

Non-Wearable None


D18 Actuating Wearable Actuation Airbag Exoskeleton Exosuit CMG None

D19 Actuating Non-Wearable Actuation

Robotic Cane Foam Pads

Mechanical Walker None

Inflatable airbags are commercially available and are triggered after a fall is successfully detected and results in the inflation of airbags in under 80ms. There are two commercially available options for inflation of airbags – pressurized CO2 canisters and sodium azide gas generators. The Hip’Safe system uses CO2 canisters while automotive airbags use gas generators containing a mixture of NaN3, KNO3, and SiO2. Both the canisters and gas generators are activated via an electrical signal. The airbags are typically made of nylon and can be folded into a very small footprint.


Lower extremity exoskeletons such as the Active Pelvis Orthosis (APO), shown in Figure 32, have been shown to improve balance and gait stability in the elderly by providing a corrective gait response within 350ms of detecting a slip [17]. This technology is still in a lab prototype stage (TRL 5 or 6) and its efficacy in mitigating falls in a real environment is currently unknown.

Exosuits, also referred to as powered clothing, fuse discreet robotics with textiles to create products that look and feel like apparel, but function more like an extension of the human body — an extra set of muscles people can put on every day. When worn by the elderly, exosuits can reduce muscle fatigue and weakness - the leading risk factors of falls. These exosuits support the body’s core by providing up to 30 watts of power to each hip and the lower back to support sitting, standing, lifting, carrying, and a range of other activities as shown in Figure 31 [71]. Seismic, a US startup, has recently launched an exosuit product line for active elders but real world data of its efficacy in mitigating falls is currently unknown. The technology can be assessed to be in the early TRL7 stage.

Figure31PowerassistprovidedbyanExosuit[72]

Control Moment Gyroscopes (CMG) are attitude control devices typically used in satellites. However, there has been recent interest in using them for addressing balance deficits and thus helping in slowing down the rate of a fall. CMG backpacks contain two reaction wheels that are capable of modifying their angular momentum to impart a moment (torque) on a body as shown in Figure 32. A relatively small gimbal motor torque and power can generate a considerably larger output moment to balance a 70kg human. The TU-Delft CMG backpack currently weights 3kgs and can support a 70kg adult [18]. Even though CMGs are mature technologies in space systems, their usage in balance support for humans is still in laboratory testing and the technology can be classified as TRL5.


Figure32WearableFallMitigationSystems(LtoR)ControlledMomentGyroscope[18],Hip'SafeAirbag[42],EPFL

Exoskeleton[17]

A non-wearable injury reduction mechanism is an assistive robot that serves as a walking cane and provides both proactive and reactive assistance to mitigate falls in the elderly user as shown in Figure 33. These types of robots are still in early concept testing and show promise in addressing balance and gait deficits in the elderly. The technology is currently in the TRL5 stage as applied to fall mitigation.

Figure33AWalkingCaneRobotasanAssistiveFallMitigationSystem[16]

Based on the review of the referenced literature, the author has rated the various injury reduction systems as shown in Table 23.


Table23RatingofInjuryReductionSystems


Actuating Wearable Airbag 1 1 5

Actuating Wearable Exoskeleton 5 5 3

Actuating Wearable Exosuit 3 3 3

Actuating Wearable CMG 3 5 1

Actuating Wearable None 0 0 0

Actuating Non-Wearable Robotic Cane 3 5 3

Actuating Non-Wearable Foam Pads 1 0 2

Actuating Non-Wearable Mechanical Walker 1 0 3

Actuating Non-Wearable None 0 0 0

4.5.6 Interacting

The interacting function is required for the user to interact with the device and to also use the device as a mechanism to interact with caregivers and emergency responders. The PERS devices and Hip’Safe have a tactile interface with a simple physical button and LED status lights. The RF sensing systems either have touchscreens on the device or use a WiFi/Bluetooth connected smartphone as the user interface. The decisions and options for a user interface is shown in Table 24.

Table24DecisionsandOptionsforInteracting/UserInterface

ID Scope Architectural Decision Alt. 1 (A1) Alt. 2 (A2) Alt. 3 (A3) Alt. 4 (A4) Alt. 5 (A5)

D16 Interacting Input Buttons Touchscreen Mic Smartphone None

D17 Interacting Output LEDs Display Speaker Smartphone None

The user survey in Chapter 3 had 85% of the respondents (n=67) indicate that they used a smartphone. A survey by the Pew Center shows that ~68% of baby boomers and ~40% of the silent generation use a smartphone (Figure 34, [73]). This suggests that a touchscreen interface or a smartphone interface will be familiar to the elderly user and could be used to provide a more interactive user experience. In recent years, voice assistants and voice recognition technology has improved significantly with services such as Alexa, Siri, Cortana and Google. There is about 20% adoption of these voices assistants in the elderly population [74] and a voice interface can help reduce the incidence of the long-lie in fall events where an automatic alert is not triggered and the person is unable to touch the device.


Figure34SmartphoneUsagebyGeneration[73]

Due to the lack of comprehensive testing data for the elderly population, the author has made best guesses for the performance rating of the various interfaces as shown in Table 25

Table25RatingofUserInterfaceSystems


Interacting Input Buttons 1 1 3

Interacting Input Touchscreen 3 5 3

Interacting Input Mic 2 3 3

Interacting Input Smartphone 1 2 3

Interacting Input None 0 0 0

Interacting Output LEDs 1 1 3

Interacting Output Display 3 3 5

Interacting Output Speaker 2 2 5

Interacting Output Smartphone 1 2 3

Interacting Output None 0 0 0

4.6 Tradespace Analysis From the previous section, each decision-option on the morphological matrix now has an associated score for cost, power consumption and performance. In addition to the GoSafe2, Hip’Safe and Kardian system concepts, an additional 22 concepts were generated through various combinations of the morphological matrix to show 5 concepts each of the Iw, Ia, Dw, Dw and Dh architecture. The charts below (Figure 35 and Figure 36) show the tradespace of Performance vs Cost and Performance vs Power for the 25 architectural concepts.


Figure35TradespaceofPerformancevsPower

Figure36TradespaceofPerformancevsCost


The Performance vs Power tradespace shows that the Iw architecture leads the pareto on the low power end and the Dh architecture leads the pareto on the high performance end. In the Performance vs Cost tradespace, the pareto front is not as clearly distinguished between the various architectures.

Table26NotablearchitecturesforFallMitigationSystems

Architecture Description Usage

Dh Distributed-Hybrid High Performance

Iw Integrated-Wearable Low Power

Based on the analysis of the reference architectures and the unmet user needs with the reference systems, a Distributed-Hybrid (Dh) architecture is recommended for modular, high-performance fall mitigations systems which facilitates the use of both wearable and ambient systems. The OPM for this architecture with the leading configuration (Dh1) is shown in Figure 37. For low-power use, low-cost use cases, the Integrated-Wearable architecture is preferable.


Figure37Dh1–DistributedHybridArchitectureOPM(boxeswithredoutlinesindicateobjectsthatarespecifictothisarchitecture)


5 Technology Roadmap

The previous chapters have covered the epidemiology of falls, user needs, system requirements, architectural decisions and identified leading architectures that could mitigate falls and reliably prevent the long-lie. The tradespace analysis illustrated the FMS performance, power consumption and cost of current technologies and their distance from the utopia point. This chapter will explore technology trends for the various fall mitigation functions, analyze technologies in the TRL 3 to 6 range and develop a technology roadmap that provides a path to move the pareto front of FMS performance, power consumption and cost closer to the utopia point.

5.1 Technology Trends A Fall Mitigation System’s four core functions: sensing, computation, powering and actuating use technologies and subsystems from the Robotics/AI field and IoT (Wearables and Smart Home) fields. Each of these fields is advancing rapidly and serve as good leading indicators for the future direction of fall mitigation systems. It is important for architects of FMS to keep an eye on the future so they can develop a strategic roadmap.

As seen in the previous chapter, power consumption is linearly correlated to performance. While there are linear performance improvements being realized with existing technologies, the focus here is on new technologies that bring a step-change improvement to FMS performance/unit power metric. Table 27 lists the technologies and their projected impact on FMS performance.

Table27ListofimpactfultechnologiesforFallMitigationSystemPerformance

Function Technology Maturity FMS Performance Improvement Impact

Communication LTE-M TRL7 Incremental

Wifi6 TRL7 Incremental

Powering Energy Harvesting – Triboelectric Nanogenerators

TRL5 Step change

Energy Harvesting – Wireless Charging TRL7 Incremental

High density Lithium chemistries TRL6 Step change

Sensing Neuromorphic MEMS sensors TRL3 Step change

Computing AI Chipsets (Tensor processing) TRL7 Incremental

Neuromorphic chips TRL5 Step change

High-efficiency Deep Learning Algorithms TRL7 Step Change


Actuating High-efficiency Exoskeletons TRL5 Step Change

Assistive Robots TRL6 Incremental

5.1.1 Neuromorphic Computing and Sensing

In recent years, robotics and artificial intelligence (AI) have placed strong demands for robust low-power sensing and for low-power, high performance computing. Neuromorphic engineering is a novel paradigm for non-Turing computation that aims to reproduce aspects of the dynamics and computational functionality found in biological brains. The goal is to replicate a brain’s hallmark functional capabilities in terms of computational power, robust learning and energy efficiency [75].

The figure of merit for computing systems is to measure the computing performance per unit power consumed (Eq. 2) and is typically referred to a GFlops/Watt or TFlops/Watt (Giga or Tera Floating Operations Per Second).

<=>?@ABCDEFGH=G>ICJF =<=>?@ABCDL?FFM

E=NFG =OP=IABCDQ?FGIAB=CREFGLFJ=CM

SIAA = TQ?RLFJ=CMSIAA =

TOP=?RSIAA

Eq.2

Figure 38 shows the performance barrier for current digital electronic systems vs neuromorphic systems and a 2019 performance comparison of processors. First generation neuromorphic chips such as TrueNorth, AIStorm and MovidiusX have already shown 100x to 1000x improvement in performance and efficiency by achieving between 1 TFlops/Watt to 10 TFlops/Watt [66]. Improvements in the fabrication processes are expected to drive further performance improvements in neuromorphic chipsets [75].

Figure38(left)Processtypesandperformance[76](right)ProcessorPerformancevsPowerconsumption,2019[66]

An FMS with a neuromorphic chipset would need 100x to 1000x lesser power budget for computation and this could open up options for running more complex machine learning (ML) algorithms onboard the device, thus further saving on the power consumption on data transfer to the cloud.


Neuromorphic techniques also have a huge application in making sensors more efficient. In current architectures, a sensor transmits raw data to a processor for computation. In neuromorphic systems, low level filtering and event classification logic can be embedded directly into the sensor, thus reducing the processing requirements on the computation system. This approach has shown great promise in vision systems where neuromorphic vision sensors have been configured to perform disparity calculations onboard to create depth images for robotics applications. Neuromorphic depth camera systems have achieved a 275x processing efficiency of 58uW/pixel while the nearest digital electronics counterpart is at 16mW/pixel [77]. By leveraging neuromorphic cameras, ambient vision sensing for FMS can be performed locally, possibly on battery powered systems without needing offboard processing in the cloud.

For wearable FMS with accelerometers, up to 50% of the power budget is for sensing (Figure 26). Recently, a neuromorphic MEMS accelerometer (micro electro-mechanical system), dubbed neuroaccelerometer, was developed with logic embedded in the sensor to classify different motion profiles. The entire pipeline of sensing to event detection was performed in the order of nanowatts vs microwatts in today’s systems [78]. A neuroaccelerometer in a wearable FMS could improve fall detection accuracy while reducing the need for high power computing and thus improving overall power efficiency by ~100x.

While neuroaccelerometers are currently at a TRL of 3, an alternative approach to improve power efficiency that is closer to TRL 7 is to use a neuromorphic chip (RAMP) as an event detector that processes raw sensor data at 10x lower power than a digital chip and wakes up the digital chip when an event such as a fall is detected [79]. See Figure 39 for the neuromorphic chip in the loop architecture. This is a viable near-term option for improving the power efficiency of wearable FMS.

Figure39Neuromorphicchipsusedaslow-powereventdetectors[79]


5.1.2 Machine Learning Algorithms

The demands from the robotics, smart home and autonomous vehicle industry have resulted in significant progress in the performance of machine learning (ML) algorithms. Coupled with the progress in processor efficiency, machine learning algorithms are being used increasingly in wearable devices. For FMS, evaluation of ML algorithms in fall detection research studies has increased from 49% in 2012 to 83+% in 2017 (Figure 40, [80]).

Figure40FallDetectionalgorithmusageinliterature(left)before2014(right)2014later[45]

Ambient sensors for FMS require quick inference from 3D datasets generated using cameras, RADAR or UWB. Deep learning algorithms have had the highest accuracy in classification of such datasets and autonomous vehicle development is further accelerating their performance. YOLOv3 is the best performing object detection algorithm but is currently optimized for GPUs (that are not power efficient. However, a fork of YOLO, called Spiking-YOLO has been demonstrated to run with 280x lower power consumption when run on neuromorphic chips such as TrueNorth [81]. Modified deep learning algorithms with neuromorphic chips can provide 100x performance/watt improvements for ambient FMS.

Figure41ComparisonofObjectDetectionAlgorithms[82]

5.1.3 Robotics

Besides airbags, alternatives to injury reduction technologies are largely in the robotics domain of wearable robots – exoskeletons and exosuits. Exoskeletons and exosuits are wearable robotic


systems that can help augment the wearer’s strength, balance and speed. The figure of merit for exoskeleton performance is the Net Metabolic Change which is defined in Eq. 3 where Edevice is the metabolic cost of walking with the device and Ebare is the metabolic cost of walking without the device.

oFApFAIq=PBJ<ℎICDF, ∆t =(t!"#$%" −t&'(")

t&'(" Eq.3

Exoskeletons have been in development for a few decades now but they only broke the metabolic cost barrier in 2013 with Malcolm et al’s active ankle exoskeleton that reduced a wearer’s metabolic cost by 6% [83]. Lim et al’s pelvic exoskeleton from 2019 is the best performing exoskeleton with a 19.8% metabolic improvement (Figure 43, [84]).

Figure42ComparisonofNetMetabolicChangeforExoskeletons[83]

Advances in materials, actuators, power sources, sensors and control systems have reduced the mass of exoskeletons and improved the control timing when power assistance is provided. For FMS, exoskeletons are at least 5 to 10 years away from general purpose use for community dwelling elders [85].


Figure43Limetal'sRoboticHipExoskeleton[84]

5.1.4 Energy Harvesting and Battery Improvements

Power consumption is a critical constraint for wearable FMS and thanks to the progress in the IoT wearables industry, both battery and energy harvesting technologies are improving at a rapid pace.

The figure of merit for batteries is the energy storage density along both volumetric and mass dimensions. It is defined as:

tCFGD{|FCRBA{ =

tCFGD{pIRR ~

Sℎ~D

�=P@>FAGBJtCFGD{|FCRBA{ =tCFGD{�=P@>F~

SℎÄ

Eq.4

It is desirable to have a battery that is both small and light with the highest energy density. The energy density of several batteries is shown in Figure 44. The ideal technology is on the top right corner of the chart with the highest volumetric and mass energy density. Thin-film Lithium metal batteries have 2x the volumetric and gravimetric energy density compared to current generation lithium chemistries and they should be available in another 2 to 3 years [86].


Figure44EnergydensityofLithiumbatterychemistries[87]

Energy harvesting from the human body and ambient sources is a viable option for powering low-power electronics. The proliferation of IoT devices – both wearable and ambient, has spurred a lot of research and development in energy harvesting from solar, RF sources, thermal, mechanical and kinetic. Figure 45 shows the various sources and energy densities of harvestable energy in a home and from a human body. Thermoelectric and kinetic energy harvesting are commercially available and have been covered in the previous chapter. A recent trend in energy harvesting is to convert mechanical friction into electricity and this is done using triboelectric nanogenerators.

Triboelectric nanogenerators or TENG are devices that effectively convert ambient mechanical energy into electricity and can be used as a power source. TENG exploit the electrification of dielectric materials when they experience friction and use the charge differential to generate a voltage. TENG has been actively researched since 2012 and when embedded in clothing, is able to generate 600uW/cm2 of power [88]. A 10cm x 10cm patch of TENG material can generate 60mW of power which is sufficient to power and charge a nominal wearable FMS. In fact, the FMS device can be packaged into TENG textile material to have a consistent power source. A TENG belt will have a surface area of 95cm x 2.5cm (CDC states average circumference for a US person is 38” or 95cm), resulting in a power generator of ~160mW.


Figure45Harvestableenergysources[89]

5.2 Roadmap Based on the above technology trends review, fall detection performance can be improved by using neuromorphic sensing and compute, deep learning algorithms, thin-film lithium batteries and triboelectric nanogenerators. Similarly, injury reduction performance can be improved by using exoskeletons. Figure 46 shows the author’s best estimate for when these technologies will reach TRL7+ and be applicable for use in Fall Mitigation Systems. The impact to FMS performance per unit power is shown in Figure 47. The biggest performance gain is from the shift to neuromorphic inertial sensing as it provides a 100x power reduction for sensing which currently has 50% of the power budget.


Figure46RoadmapoftechnologyjumpsforFMS

Figure47FMSPerformance/UnitPowerRoadmap

There are several ways to use these roadmaps to inform technology and development strategy. For a company that is primarily focused on wearables and fall mitigation systems, Figure 47 can serve as an investment prioritization guideline. Given that neuromorphic inertial sensing has the highest performance gain, one could consider options for accelerating its development and technology maturity timelines so that it can be used in FMS earlier. Additionally, this timeline could also be used to gain a competitive advantage by choosing to invest in certain core technologies and patent them to lock out the competition.


6 Conclusion

This thesis used a systems approach to analyze different architectures for a fall mitigation system architecture that can detect falls, reduce injury and issue emergency alerts to reliably prevent the long-lie in independent elders.

In Chapter 2, a National Health Interview Survey data was analyzed to understand the causes for falls, types of fall related injuries and common fall locations for community dwelling elders. Weakness, gait deficit and balance deficit are the top 3 leading risk factors for falls and fractures to the arms, legs and hips are the injuries to protect.

A concept of operations was defined based on these findings and in Chapter 3, a user survey was conducted to understand the needs of community dwelling elders and the results were analyzed to prioritize system requirements for a fall mitigation system (FMS). The survey respondents indicated a strong preference for a discreet, always-on wearable device that works both indoors and outdoors while indicating a lower preference for injury reduction functions. The author believes the lower preference for injury reduction was either due to a misunderstanding of the survey question or due to a lack of familiarity of available injury reduction options.

In Chapter 4, An FMS was decomposed into six level 2 functions and the various form choices for each of these functions were analyzed and rated for performance, power consumption and cost. Five different fall mitigation system architectures were analyzed and the Distributed-Hybrid architecture had the highest performance while the Integrated-Wearable architecture had the lowest power consumption.

In Chapter 5, future technology trends in robotics, AI, neuromorphic computing and energy harvesting were studied to create a long-term strategic roadmap for fall mitigation systems. Neuromorphic systems for sensing and computing offer the biggest performance per unit power benefit and enable other performance improvements with the ability to run deep learning algorithms locally. Investments should be prioritized for neuromorphic architectures to accelerate their time to market. Exoskeletons have the longest lead time but have a noticeable impact to performance due to their ability to address the top 3 risk factors for falls.

Limitations

There are several gaps in this thesis that future researchers should be aware of. Most of the fall related datasets are from US sources and the results might not be representative of global challenges. The user survey is from a small sample size of 79 respondents and most of them are healthy and independent with no prior history of falls. Before proceeding with any detailed design activity, it is important to conduct surveys of current users of fall mitigation systems to gain a better understanding of their challenges.

The thesis focused explicitly on user needs and made implicit assumptions about the needs of the other stakeholders (healthcare providers, insurance, family). A more detailed and explicit analysis of the other stakeholders is required to form a complete picture of the system requirements for fall mitigation systems.


The thesis is primarily focused on non-fatal falls and makes an implicit assumption that automatic alerting to an emergency control center will guarantee that a user will receive timely help and avoid the long-lie. While this might be a reasonable assumption for users in urban settings in the US where emergency responders arrive within 15 mins, it is not a good assumption for rural areas or for countries that are in the early stages of developing their health care systems. The architectural preferences and recommendations will change from a Distributed-Hybrid system to a low-power, low-cost Integrated-wearable system for a user base in a less developed country.


7 References

[1] World Health Organization, Ed., WHO global report on falls prevention in older age. Geneva, Switzerland: World Health Organization, 2008.

[2] “World Population Prospects - Population Division - United Nations.” https://population.un.org/wpp/Graphs/Probabilistic/POP/65plus/900 (accessed Apr. 28, 2020).

[3] “Fact Sheet: Aging in the United States – Population Reference Bureau.” https://www.prb.org/aging-unitedstates-fact-sheet/ (accessed Apr. 29, 2020).

[4] “Fact Sheet: Falls | Aging.com.” https://www.aging.com/falls-fact-sheet/ (accessed Dec. 17, 2019).

[5] C. S. Florence, G. Bergen, A. Atherly, E. Burns, J. Stevens, and C. Drake, “Medical Costs of Fatal and Nonfatal Falls in Older Adults,” J. Am. Geriatr. Soc., vol. 66, no. 4, pp. 693–698, 2018, doi: 10.1111/jgs.15304.

[6] Y. Dionyssiotis, “Analyzing the problem of falls among older people,” Int. J. Gen. Med., vol. 5, pp. 805–813, Sep. 2012, doi: 10.2147/IJGM.S32651.

[7] S. Schnell, S. M. Friedman, D. A. Mendelson, K. W. Bingham, and S. L. Kates, “The 1-Year Mortality of Patients Treated in a Hip Fracture Program for Elders,” Geriatr. Orthop. Surg. Rehabil., vol. 1, no. 1, pp. 6–14, Sep. 2010, doi: 10.1177/2151458510378105.

[8] M. P. Phy et al., “Effects of a hospitalist model on elderly patients with hip fracture,” Arch. Intern. Med., vol. 165, no. 7, pp. 796–801, Apr. 2005, doi: 10.1001/archinte.165.7.796.

[9] N. C. Voermans, A. H. Snijders, Y. Schoon, and B. R. Bloem, “Why old people fall (and how to stop them),” Pract. Neurol., vol. 7, no. 3, pp. 158–171, Jun. 2007, doi: 10.1136/jnnp.2007.120980.

[10] Accius, Jean, and Joo Yeoun Suh., “The Longevity Economy Outlook: How People Ages 50 and Older Are Fueling Economic Growth, Stimulating Jobs, and Creating Opportunities for All.,” AARP Thought Leadersh., Dec. 2019, doi: https://doi.org/10.26419/int.00042.001.

[11] C. Sherrington et al., “Exercise for preventing falls in older people living in the community,” Cochrane Database Syst. Rev., vol. 2019, no. 1, Jan. 2019, doi: 10.1002/14651858.CD012424.pub2.

[12] T. Smits and A. Ryter, “Personal Emergency Response System (PERS) with Optimized Automatic Fall Detection Shows Greater Effectiveness than PERS Alone.” Philips Lifeline, [Online]. Available: https://www.lifeline.philips.com/content/dam/PLL/PLL-Common/PDFs/WhitePaper-AA_Analytics-j.pdf.

[13] R. Stokke, “The Personal Emergency Response System as a Technology Innovation in Primary Health Care Services: An Integrative Review,” J. Med. Internet Res., vol. 18, no. 7, Jul. 2016, doi: 10.2196/jmir.5727.

[14] J. Fleming and C. Brayne, “Inability to get up after falling, subsequent time on floor, and summoning help: prospective cohort study in people over 90,” The BMJ, vol. 337, Nov. 2008, doi: 10.1136/bmj.a2227.

[15] “Study Shows Vibrating Insoles Could Reduce Falls Among Seniors,” Wyss Institute, Oct. 28, 2014. https://wyss.harvard.edu/news/study-shows-vibrating-insoles-could-reduce-falls-among-seniors/ (accessed Dec. 18, 2019).


[16] P. Di et al., “Fall Detection and Prevention Control Using Walking-Aid Cane Robot,” IEEEASME Trans. Mechatron., vol. 21, no. 2, pp. 625–637, Apr. 2016, doi: 10.1109/TMECH.2015.2477996.

[17] V. Monaco et al., “An ecologically-controlled exoskeleton can improve balance recovery after slippage,” Sci. Rep., vol. 7, no. 1, pp. 1–10, May 2017, doi: 10.1038/srep46721.

[18] D. Lemus, J. van Frankenhuyzen, and H. Vallery, “Design and Evaluation of a Balance Assistance Control Moment Gyroscope,” J. Mech. Robot., vol. 9, no. 5, Oct. 2017, doi: 10.1115/1.4037255.

[19] “Walabot Fall Alert System | Detect Falls with No Wearables,” Walabot Official Site | Walabot 3D imaging - Uncover the hidden world around you. https://walabot.com/walabot-home (accessed Mar. 29, 2020).

[20] “How fall detection is moving beyond the pendant,” MobiHealthNews, Apr. 19, 2019. https://www.mobihealthnews.com/content/how-fall-detection-moving-beyond-pendant (accessed Mar. 28, 2020).

[21] MITRE Systems Engineering Guide. MITRE Corporation, 2012. [22] “Chapter 1. What is epidemiology? | The BMJ.” https://www.bmj.com/about-

bmj/resources-readers/publications/epidemiology-uninitiated/1-what-epidemiology (accessed Mar. 22, 2020).

[23] “NHIS - About the National Health Interview Survey,” Jun. 28, 2019. https://www.cdc.gov/nchs/nhis/about_nhis.htm (accessed Mar. 22, 2020).

[24] S. K. Verma, J. L. Willetts, H. L. Corns, H. R. Marucci-Wellman, D. A. Lombardi, and T. K. Courtney, “Falls and Fall-Related Injuries among Community-Dwelling Adults in the United States,” PLoS ONE, vol. 11, no. 3, Mar. 2016, doi: 10.1371/journal.pone.0150939.

[25] S. Yoshida, “A Global Report on Falls Prevention Epidemiology of Falls.” World Health Organization.

[26] L. Z. Rubenstein, “Falls in older people: epidemiology, risk factors and strategies for prevention,” Age Ageing, vol. 35, no. suppl_2, pp. ii37–ii41, Sep. 2006, doi: 10.1093/ageing/afl084.

[27] T. Masud and R. O. Morris, “Epidemiology of falls,” Age Ageing, vol. 30, no. suppl 4, pp. 3–7, Nov. 2001, doi: 10.1093/ageing/30.suppl_4.3.

[28] F. Eto, “Causes of Falls in the Elderly,” p. 7. [29] L. Min, M. Tinetti, K. M. Langa, J. Ha, N. Alexander, and G. J. Hoffman, “Measurement of

Fall Injury With Health Care System Data and Assessment of Inclusiveness and Validity of Measurement Models,” JAMA Netw. Open, vol. 2, no. 8, pp. e199679–e199679, Aug. 2019, doi: 10.1001/jamanetworkopen.2019.9679.

[30] A. Bergland, G.-B. Jarnlo, and K. Laake, “Predictors of falls in the elderly by location,” Aging Clin. Exp. Res., vol. 15, no. 1, pp. 43–50, Feb. 2003, doi: 10.1007/BF03324479.

[31] “ISO - Foreword - Supplementary information,” ISO. https://www.iso.org/foreword-supplementary-information.html (accessed Apr. 06, 2020).

[32] 14:00-17:00, “ISO/IEC/IEEE 15288:2015,” ISO. https://www.iso.org/cms/render/live/en/sites/isoorg/contents/data/standard/06/37/63711.html (accessed Apr. 06, 2020).


[33] E. Crawley, B. Cameron, and D. Selva, System Architecture: Strategy and Product Development for Complex Systems, 1st ed. USA: Prentice Hall Press, 2015.

[34] “UnitedHealthcare | Philips Lifeline ®,” Philips. https://www.lifeline.philips.com/uhcmedicare.html (accessed May 04, 2020).

[35] “GoSafe 2 | Philips Lifeline ®,” Philips. https://www.lifeline.philips.com/medical-alert-systems/gosafe-2.html (accessed Apr. 11, 2020).

[36] “CFR - Code of Federal Regulations Title 21.” https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfcfr/cfrsearch.cfm?fr=890.3710 (accessed Apr. 04, 2020).

[37] “CFR - Code of Federal Regulations Title 21.” https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfcfr/cfrsearch.cfm?fr=890.3725 (accessed Apr. 04, 2020).

[38] Q. Zhang, H. Q. Li, Y. K. Ning, D. Liang, and G. R. Zhao, “Design and Realization of a Wearable Hip-Airbag System for Fall Protection,” Appl. Mech. Mater., vol. 461, pp. 667–674, Nov. 2013, doi: 10.4028/www.scientific.net/AMM.461.667.

[39] O. Aziz, M. Musngi, E. J. Park, G. Mori, and S. N. Robinovitch, “A comparison of accuracy of fall detection algorithms (threshold-based vs. machine learning) using waist-mounted tri-axial accelerometer signals from a comprehensive set of falls and non-fall trials,” Med. Biol. Eng. Comput., vol. 55, no. 1, pp. 45–55, Jan. 2017, doi: 10.1007/s11517-016-1504-y.

[40] “Automatic Fall Detection | Philips Lifeline ®,” Philips. https://www.lifeline.philips.com/medical-alert-systems/fall-detection.html (accessed Apr. 11, 2020).

[41] “Hip protection - prevent fractures with hip protectors |.” https://tytex.com/products/hip-protection (accessed Apr. 11, 2020).

[42] “Hip’Safe by Helite – Airbag protection for seniors.” https://senior.helite.com/en/ (accessed Dec. 18, 2019).

[43] S. Duma et al., “DEVELOPMENT OF AN INFLATABLE HIP PROTECTION SYSTEM: DESIGN FOR HIP FRACTURE PREVENTION AND INCREASED COMPLIANCE,” p. 2.

[44] “How It Works,” Tango Belt - Hip Protection Redefined. https://www.tangobelt.com/how-it-works (accessed Apr. 18, 2020).

[45] T. Xu, Y. Zhou, and J. Zhu, “New Advances and Challenges of Fall Detection Systems: A Survey,” Appl. Sci., vol. 8, no. 3, p. 418, Mar. 2018, doi: 10.3390/app8030418.

[46] L. Almazaydeh, K. Al-Otoon, A. Al-Dmour, K. Elleithy, and K. Alatoun, “A Panoramic Study of Fall Detection Technologies,” Int. J. Comput. Sci. Issues, vol. 13, pp. 62–65, May 2016, doi: 10.20943/01201603.6265.

[47] Y. S. Delahoz and M. A. Labrador, “Survey on Fall Detection and Fall Prevention Using Wearable and External Sensors,” Sensors, vol. 14, no. 10, pp. 19806–19842, Oct. 2014, doi: 10.3390/s141019806.

[48] A. Ramachandran and A. Karuppiah, “A Survey on Recent Advances in Wearable Fall Detection Systems,” BioMed Research International, 2020. https://www.hindawi.com/journals/bmri/2020/2167160/ (accessed Mar. 14, 2020).

[49] M. Ahmed, N. Mehmood, A. Nadeem, A. Mehmood, and K. Rizwan, “Fall Detection System for the Elderly Based on the Classification of Shimmer Sensor Prototype Data,” Healthc. Inform. Res., vol. 23, no. 3, pp. 147–158, Jul. 2017, doi: 10.4258/hir.2017.23.3.147.


[50] “Ambient Sensors for Elderly Care and Independent Living: A Survey. - Abstract - Europe PMC.” https://europepmc.org/article/PMC/6068532 (accessed Apr. 19, 2020).

[51] D. Ajerla, S. Mahfuz, and F. Zulkernine, “A Real-Time Patient Monitoring Framework for Fall Detection,” Wireless Communications and Mobile Computing, 2019. https://www.hindawi.com/journals/wcmc/2019/9507938/ (accessed Apr. 19, 2020).

[52] “Different Ways to Connect IoT Devices to Transmit and Receive Data,” Embedded AMS, Oct. 15, 2019. https://embeddedams.nl/different-ways-to-connect-iot-devices-to-transmit-and-receive-data/ (accessed Apr. 12, 2020).

[53] C. Culman, S. Aminikhanghahi, and D. J. Cook, “Easing Power Consumption of Wearable Activity Monitoring with Change Point Detection,” Sensors, vol. 20, no. 1, Jan. 2020, doi: 10.3390/s20010310.

[54] M. Keenan and M. Keenan, “Supercapacitor And Li-Ion Technology Combining | Avnet Abacus.” https://www.avnet.com/wps/portal/abacus/resources/article/hybrid-capacitors-combine-supercacitor-and-li-on-technology (accessed Apr. 27, 2020).

[55] L. Mearian, “Wireless charging explained: What is it and how does it work?,” Computerworld, Mar. 28, 2018. https://www.computerworld.com/article/3235176/wireless-charging-explained-what-is-it-and-how-does-it-work.html (accessed Apr. 27, 2020).

[56] “The State of Wireless Charging in 2019 and Beyond,” Digital Trends, Feb. 11, 2019. https://www.digitaltrends.com/mobile/state-of-wireless-charging-2019/ (accessed Apr. 27, 2020).

[57] “Energy Harvesting — MATRIX Industries.” https://www.matrixindustries.com/en/energy-harvesting (accessed Apr. 12, 2020).

[58] “(PDF) Wearable Thermoelectric Generators for Body-Powered Devices,” ResearchGate. https://www.researchgate.net/publication/227317323_Wearable_Thermoelectric_Generators_for_Body-Powered_Devices (accessed Apr. 11, 2020).

[59] “Micro Generator Technology,” Kinetron. https://www.kinetron.eu/smart-generators/ (accessed Apr. 27, 2020).

[60] “Sensor,” Wikipedia. Apr. 03, 2020, Accessed: Apr. 11, 2020. [Online]. Available: https://en.wikipedia.org/w/index.php?title=Sensor&oldid=948856845.

[61] W. Lu, M. C. Stevens, C. Wang, S. J. Redmond, and N. H. Lovell, “Smart Triggering of the Barometer in a Fall Detector Using a Semi-Permeable Membrane,” IEEE Trans. Biomed. Eng., vol. 67, no. 1, pp. 146–157, Jan. 2020, doi: 10.1109/TBME.2019.2909907.

[62] M. Tolkiehn, L. Atallah, B. Lo, and G.-Z. Yang, “Direction sensitive fall detection using a triaxial accelerometer and a barometric pressure sensor,” Conf. Proc. Annu. Int. Conf. IEEE Eng. Med. Biol. Soc. IEEE Eng. Med. Biol. Soc. Annu. Conf., vol. 2011, pp. 369–372, 2011, doi: 10.1109/IEMBS.2011.6090120.

[63] “Wearable Weight Loss - Slim Down Wellness Devices with Better Energy Efficiency.” https://www.maximintegrated.com/en/design/technical-documents/app-notes/5/5949.html (accessed Apr. 27, 2020).

[64] “(PDF) Gazelle: Energy-Efficient Wearable Analysis for Running,” ResearchGate. https://www.researchgate.net/publication/309591381_Gazelle_Energy-Efficient_Wearable_Analysis_for_Running (accessed Apr. 11, 2020).


[65] G. Aroganam, N. Manivannan, and D. Harrison, “Review on Wearable Technology Sensors Used in Consumer Sport Applications,” Sensors, vol. 19, no. 9, Apr. 2019, doi: 10.3390/s19091983.

[66] A. Reuther, P. Michaleas, M. Jones, V. Gadepally, S. Samsi, and J. Kepner, “Survey and Benchmarking of Machine Learning Accelerators,” 2019 IEEE High Perform. Extreme Comput. Conf. HPEC, pp. 1–9, Sep. 2019, doi: 10.1109/HPEC.2019.8916327.

[67] “AI on a MEMS Device Brings Neuromorphic Computing to the Edge - IEEE Spectrum,” IEEE Spectrum: Technology, Engineering, and Science News. https://spectrum.ieee.org/tech-talk/artificial-intelligence/embedded-ai/artificial-intelligence-on-a-mems-device-brings-neuromorphic-computing-to-the-edge (accessed Apr. 24, 2020).

[68] L. Ren and Y. Peng, “Research of Fall Detection and Fall Prevention Technologies: A Systematic Review,” IEEE Access, vol. 7, pp. 77702–77722, 2019, doi: 10.1109/ACCESS.2019.2922708.

[69] A. T. Özdemir, “An Analysis on Sensor Locations of the Human Body for Wearable Fall Detection Devices: Principles and Practice,” Sensors, vol. 16, no. 8, p. 1161, Aug. 2016, doi: 10.3390/s16081161.

[70] T. Øvrebekk, “The Importance of Average Power Consumption to Battery Life.” https://blog.nordicsemi.com/getconnected/the-importance-of-average-power-consumption-to-battery-life (accessed Apr. 11, 2020).

[71] S. Francisco, “Seismic Unveils First Collection of Powered ClothingT M,” p. 5. [72] “Posts Archive,” fuseproject. https://fuseproject.com/blog (accessed Apr. 26, 2020). [73] 1615 L. St NW, Suite 800Washington, and D. 20036USA202-419-4300 | M.-857-8562 | F.-

419-4372 | M. Inquiries, “Millennials stand out for their technology use, but older generations also embrace digital life,” Pew Research Center. https://www.pewresearch.org/fact-tank/2019/09/09/us-generations-technology-use/ (accessed Apr. 27, 2020).

[74] “Voice Assistant Demographic Data - Young Consumers More Likely to Own Smart Speakers While Over 60 Bias Toward Alexa and Siri,” Voicebot.ai, Jun. 21, 2019. https://voicebot.ai/2019/06/21/voice-assistant-demographic-data-young-consumers-more-likely-to-own-smart-speakers-while-over-60-bias-toward-alexa-and-siri/ (accessed Apr. 27, 2020).

[75] T. Wunderlich et al., “Demonstrating Advantages of Neuromorphic Computation: A Pilot Study,” Front. Neurosci., vol. 13, 2019, doi: 10.3389/fnins.2019.00260.

[76] T. F. de Lima, B. J. Shastri, A. N. Tait, M. A. Nahmias, and P. R. Prucnal, “Progress in neuromorphic photonics,” Nanophotonics, vol. 6, no. 3, pp. 577–599, Mar. 2017, doi: 10.1515/nanoph-2016-0139.

[77] A. Andreopoulos, H. J. Kashyap, T. K. Nayak, A. Amir, and M. D. Flickner, “A Low Power, High Throughput, Fully Event-Based Stereo System,” in 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, Jun. 2018, pp. 7532–7542, doi: 10.1109/CVPR.2018.00786.

[78] B. Barazani, G. Dion, J.-F. Morissette, L. Beaudoin, and J. Sylvestre, “Microfabricated Neuroaccelerometer: Integrating Sensing and Reservoir Computing in MEMS,” J. Microelectromechanical Syst., vol. PP, pp. 1–10, Mar. 2020, doi: 10.1109/JMEMS.2020.2978467.


[79] “Power- and Data-Efficient Edge Processor | Aspinity.” https://www.aspinity.com/Technology (accessed May 01, 2020).

[80] “(PDF) A Comparison of Machine Learning Algorithms for Fall Detection using Wearable Sensors,” ResearchGate. https://www.researchgate.net/publication/339414213_A_Comparison_of_Machine_Learning_Algorithms_for_Fall_Detection_using_Wearable_Sensors (accessed Apr. 19, 2020).

[81] “Spiking-YOLO: Spiking Neural Network for Energy-Efficient Object Detection,” GroundAI. https://www.groundai.com/project/spiking-yolo-spiking-neural-network-for-energy-efficient-object-detection/ (accessed May 01, 2020).

[82] J. Redmon and A. Farhadi, “YOLOv3: An Incremental Improvement,” p. 6. [83] G. S. Sawicki, O. N. Beck, I. Kang, and A. J. Young, “The exoskeleton expansion: improving

walking and running economy,” J. NeuroEngineering Rehabil., vol. 17, no. 1, p. 25, Feb. 2020, doi: 10.1186/s12984-020-00663-9.

[84] B. Lim et al., “Delayed Output Feedback Control for Gait Assistance With a Robotic Hip Exoskeleton,” IEEE Trans. Robot., vol. 35, no. 4, pp. 1055–1062, Aug. 2019, doi: 10.1109/TRO.2019.2913318.

[85] R. Rajagopalan, I. Litvan, and T.-P. Jung, “Fall Prediction and Prevention Systems: Recent Trends, Challenges, and Future Research Directions,” Sensors, vol. 17, no. 11, Nov. 2017, doi: 10.3390/s17112509.

[86] “(PDF) Solid-State Thin-Film Lithium Batteries for Integration in Microsystems,” ResearchGate. https://www.researchgate.net/publication/235407571_Solid-State_Thin-Film_Lithium_Batteries_for_Integration_in_Microsystems (accessed Apr. 12, 2020).

[87] Y. Liang et al., “A review of rechargeable batteries for portable electronic devices,” InfoMat, vol. 1, no. 1, pp. 6–32, 2019, doi: 10.1002/inf2.12000.

[88] C. Wu, A. C. Wang, W. Ding, H. Guo, and Z. L. Wang, “Triboelectric Nanogenerator: A Foundation of the Energy for the New Era,” Adv. Energy Mater., vol. 9, no. 1, p. 1802906, 2019, doi: 10.1002/aenm.201802906.

[89] “(2) (PDF) Towards the self-powered Internet of Things (IoT) by energy harvesting: Trends and technologies for green IoT,” ResearchGate. https://www.researchgate.net/publication/324997344_Towards_the_self-powered_Internet_of_Things_IoT_by_energy_harvesting_Trends_and_technologies_for_green_IoT (accessed Apr. 30, 2020).


8 Appendix A: User Requirements Survey


9 Appendix B: Tradespace Analysis

Integrated-Wearable (Iw)

ID Scope

Architectural Decision

Alt. 1 (A1)

Alt. 2 (A2)

Alt. 3 (A3)

Alt. 4 (A4)

Alt. 5 (A5)

Iw1-PERS

Iw2-HipSafe Iw3 Iw4 Iw5

D1

Glob

al

Sensing Mode

Wearab

le Ambient

Hybrid

Sens

Wearab

le

Weara

ble

Wearabl

e

Wearab

le

Wearab

le

D2

Glob

al

Injury Reduction Mode

Wearab

le Inj

Non-

Wearabl

e None None

Weara

ble None

Wearab

le Inj None

D3

Glob

al

Computing Mode

Onboar

d

Offboar

d

Hybrid

Comp None

Onboar

d

Onboa

rd

Onboar

d

Onboar

d

Onboar

d

D4

Glob

al

Number of devices Single Multiple Single Single Single Single Single

D5

Glob

al

Communication Mode

Short-

Range

Long-

Range

Hybrid

Comms None

Long-

Range None

Long-

Range

Hybrid

Comms

Short-

Range

D6

Glob

al Alerting Local

Control

Center

Hybrid

Alert None

Control

Center None

Control

Center

Hybrid

Alert

Hybrid

Alert

D7

Sens

ing

Wearable Sensing

Acceler

ometer

Gyrosco

pe +

Acc.

Baromet

er + Acc. None

Barome

ter +

Acc.

Acceler

ometer

Baromet

er + Acc.

Gyrosco

pe +

Acc.

Barome

ter +

Acc.

D8

Sens

ing

Ambient Sensing

mmWa

ve

RADAR UWB Camera IR

Non

e None None None None None

D9

Pow

erin

g Storage Lithium

Battery

Superca

pacitor None

Lithium

Battery

Lithiu

m

Battery

Lithium

Battery

Lithium

Battery

Lithium

Battery

D10

Pow

erin

g Recharging TEG

Wireless

Chargin

g Wall

Kineti

c

Non

e Wall Wall

Wireless

Chargin

g Wall TEG

D11

Com

mun

icati

ng

Short-Range BLE ANT+ Zigbee Wifi

Non

e None None None BLE BLE

D12

Com

mun

icati

ng Long-Range LTE-M LoRA SigFox 4G

Non

e 4G None LTE-M LTE-M None

D13

Com

puti

ng Onboard Tensor

Chip

Mobile

chip FPGA ASIC

Non

e

Mobile

chip

Mobile

chip ASIC

Mobile

chip ASIC

D14

Com

puti

ng Offboard Cloud Edge None None None None None None

D15

Com

puti

ng Algorithms Thresho

ld

ML -

kNN/SV

M

Deep

Learning None

Thresho

ld

Thresh

old

ML -

kNN/SV

M

ML -

kNN/SV

M

ML -

kNN/SV

M

D16

Inte

racti

ng Input Buttons

Touchsc

reen Mic

Smart

phon

e

Non

e Buttons

Button

s Buttons

Touchsc

reen

Smartp

hone

D17

Inte

racti

ng Output LEDs Display Speaker

Smart

phon

e

Non

e LEDs LEDs LEDs Display

Smartp

hone

D18

Actu

atin

g

Wearable Actuation Airbag

Exoskele

ton Exosuit CMG

Non

e None Airbag None Exosuit None

D19

Actu

atin

g

Non-Wearable Actuation

Robotic

Cane

Foam

Pads

Mechani

cal

Walker None None None None None None


Integrated-Ambient (Ia)

ID Scope Architectural Decision

Alt. 1 (A1)

Alt. 2 (A2)

Alt. 3 (A3)

Alt. 4 (A4)

Alt. 5 (A5) Ia1 Ia2 Ia3 Ia4 Ia5

D1 Global

Sensing Mode

Wearab

le

Ambien

t

Hybrid

Sens

Ambie

nt

Ambie

nt

Ambien

t

Ambie

nt Ambient

D2 Global


Wearab

le Inj

Non-

Wearabl

e None None None None

Non-

Weara

ble

Non-

Wearabl

e

D3 Global

Computing Mode

Onboar

d

Offboar

d

Hybrid

Comp None

Onboa

rd

Onboa

rd

Onboar

d

Onboa

rd Onboard

D4 Global

Number of devices Single Multiple Single Single Single Single Single

D5 Global

Communication Mode

Short-

Range

Long-

Range

Hybrid

Comms None

Short-

Range

Short-

Range

Hybrid

Comms

Hybrid

Comm

s

Short-

Range

D6 Global Alerting Local

Control

Center

Hybrid

Alert None

Hybrid

Alert

Hybrid

Alert

Hybrid

Alert

Hybrid

Alert

Hybrid

Alert

D7

Sensin

g

Wearable Sensing

Acceler

ometer

Gyrosco

pe +

Acc.

Baromet

er + Acc. None None None None None None

D8

Sensin

g

Ambient Sensing

mmWa

ve

RADAR UWB Camera IR

Non

e UWB

Camer

a

mmWa

ve

RADAR None IR

D9

Poweri

ng Storage Lithium

Battery

Superca

pacitor None

Lithiu

m

Battery

Lithiu

m

Battery

Lithium

Battery

Lithiu

m

Battery None

D10

Poweri

ng Recharging TEG

Wireless

Chargin

g Wall

Kineti

c

Non

e Wall Wall Wall Wall Wall

D11

Comm

unicati

ng


Non

e Wifi Wifi Wifi Wifi Wifi

D12

Comm

unicati


Non

e None None 4G 4G None

D13

Compu

ting Onboard Tensor

Chip

Mobile

chip FPGA ASIC

Non

e

Mobile

chip

Tensor

Chip

Tensor

Chip

Mobile

chip FPGA

D14

Compu

ting Offboard Cloud Edge None None None None None None

D15

Compu

ting Algorithms Thresho

ld

ML -

kNN/SV

M

Deep

Learning None

ML -

kNN/S

VM

Deep

Learni

ng

ML -

kNN/SV

M

Thresh

old

ML -

kNN/SV

M

D16

Interac

ting Input Buttons

Touchsc

reen Mic

Smar

tpho

ne

Non

e

Smartp

hone

Smartp

hone

Smartp

hone

Touchs

creen Mic

D17

Interac

ting Output LEDs Display Speaker

Smar

tpho

ne

Non

e

Smartp

hone

Smartp

hone

Smartp

hone Display Speaker

D18

Actuati

ng


Exoskel

eton Exosuit CMG

Non


D19

Actuati

ng


Robotic

Cane

Foam

Pads

Mechani

cal

Walker None None None None

Roboti

c Cane

Mechani

cal

Walker


Distributed-Ambient (Da)


Alt. 1 (A1)

Alt. 2 (A2)

Alt. 3 (A3)

Alt. 4 (A4)

Alt. 5 (A5)

Da1-Kardian Da2 Da3 Da4 Da5

D1 Global

Sensing Mode

Wearab

le

Ambien

t

Hybrid

Sens

Ambien

t

Ambien

t

Ambien

t

Ambie

nt

Ambien

t

D2 Global


Wearab

le Inj

Non-

Wearabl

e None None

Non-

Wearabl

e None None None

D3 Global

Computing Mode

Onboar

d

Offboar

d

Hybrid

Comp None

Onboar

d

Hybrid

Comp

Onboar

d

Onboa

rd

Onboar

d

D4 Global

Number of devices Single Multiple

Multipl

e Multiple

Multipl

e

Multip

le Multiple

D5 Global

Communication Mode

Short-

Range

Long-

Range

Hybrid

Comms None

Short-

Range

Hybrid

Comms

Hybrid

Comms

Short-

Range

Short-

Range


Control

Center

Hybrid

Alert None

Hybrid

Alert

Hybrid

Alert

Hybrid

Alert Local None

D7

Sensin

g

Wearable Sensing

Acceler

ometer

Gyrosco

pe +

Acc.

Baromet

er + Acc. None None None None None None

D8

Sensin

g

Ambient Sensing

mmWa

ve

RADAR UWB Camera IR

Non

e

mmWa

ve

RADAR

mmWav

e

RADAR

mmWa

ve

RADAR

Camer

a UWB

D9

Poweri

ng Storage Lithium

Battery

Superca

pacitor None None

Lithium

Battery

Lithium

Battery None

Lithium

Battery

D10

Poweri

ng Recharging TEG

Wireles

s

Chargin

g Wall

Kineti

c

Non

e Wall

Wireles

s

Chargin

g Wall Wall

Wireles

s

Chargin

g

D11

Comm

unicati

ng


Non

e Wifi BLE Wifi Wifi BLE

D12

Comm

unicati


Non

e None LTE-M LTE-M None None

D13

Compu

ting Onboard Tensor

Chip

Mobile

chip FPGA ASIC

Non

e

Tensor

Chip

Tensor

Chip

Tensor

Chip

Tensor

Chip

Mobile

chip

D14

Compu

ting Offboard Cloud Edge None None Cloud None None None

D15

Compu

ting Algorithms Thresho

ld

ML -

kNN/SV

M

Deep

Learning None

Deep

Learnin

g

ML -

kNN/SV

M

ML -

kNN/SV

M

Deep

Learni

ng

Thresho

ld

D16

Interac

ting Input Buttons

Touchsc

reen Mic

Smar

tpho

ne

Non

e

Touchsc

reen

Smartp

hone

Smartp

hone

Smart

phone Buttons

D17

Interac


Smar

tpho

ne

Non

e Display

Smartp

hone

Smartp

hone

Smart

phone LEDs

D18

Actuati

ng


Exoskel

eton Exosuit CMG

Non


D19

Actuati

ng


Robotic

Cane

Foam

Pads

Mechani

cal

Walker None None

Robotic

Cane None None None


Distributed-Wearable (Dw)


Alt. 1 (A1)

Alt. 2 (A2)

Alt. 3 (A3)

Alt. 4 (A4)

Alt. 5 (A5) Dw1 Dw2 Dw3 Dw4 Dw5

D1 Global

Sensing Mode

Wearab

le

Ambien

t

Hybrid

Sens

Wearab

le

Wearab

le

Wearab

le

Wearab

le

Wearab

le

D2 Global


Wearab

le Inj

Non-

Wearab

le None

Wearab

le Inj

Wearab

le Inj

Wearab

le Inj

Wearab

le Inj None

D3 Global

Computing Mode

Onboar

d

Offboar

d

Hybrid

Comp None

Onboar

d

Hybrid

Comp

Onboar

d

Hybrid

Comp

Offboar

d

D4 Global

Number of devices Single

Multipl

e

Multipl

e

Multipl

e

Multipl

e

Multipl

e

Multipl

e

D5 Global

Communication Mode

Short-

Range

Long-

Range

Hybrid

Comms None

Short-

Range

Long-

Range

Hybrid

Comms

Hybrid

Comms

Short-

Range


Control

Center

Hybrid

Alert None

Hybrid

Alert

Hybrid

Alert

Hybrid

Alert

Hybrid

Alert

Hybrid

Alert

D7

Sensin

g

Wearable Sensing

Acceler

ometer

Gyrosco

pe +

Acc.

Baromet

er + Acc. None

Barome

ter +

Acc.

Gyrosco

pe +

Acc.

Gyrosco

pe +

Acc.

Gyrosc

ope +

Acc.

Barome

ter +

Acc.

D8

Sensin

g

Ambient Sensing

mmWa

ve

RADAR UWB Camera IR

Non


D9

Poweri

ng Storage Lithium

Battery

Superca

pacitor None

Lithium

Battery

Lithium

Battery

Lithium

Battery

Lithium

Battery

Superca

pacitor

D10

Poweri

ng Recharging TEG

Wireles

s

Chargin

g Wall

Kineti

c

Non

e TEG

Wireles

s

Chargin

g

Wireles

s

Chargin

g Wall TEG

D11

Comm

unicati

ng


Non

e BLE None BLE BLE BLE

D12

Comm

unicati


Non

e None LTE-M LTE-M LTE-M None

D13

Compu

ting Onboard Tensor

Chip

Mobile

chip FPGA ASIC

Non

e

Tensor

Chip ASIC

Tensor

Chip

Mobile

chip ASIC

D14

Compu

ting Offboard Cloud Edge None None Cloud None Cloud Cloud

D15

Compu

ting Algorithms Thresh

old

ML -

kNN/SV

M

Deep

Learning None

ML -

kNN/SV

M

ML -

kNN/SV

M

ML -

kNN/SV

M

ML -

kNN/SV

M

ML -

kNN/SV

M

D16

Interac

ting Input Buttons

Touchsc

reen Mic

Smar

tpho

ne

Non

e

Smartp

hone

Touchsc

reen Buttons

Touchs

creen

Smartp

hone

D17

Interac


Smar

tpho

ne

Non

e

Smartp

hone Display LEDs Display

Smartp

hone

D18

Actuati

ng


Exoskel

eton Exosuit CMG

Non

e Airbag

Exoskel

eton Exosuit CMG None

D19

Actuati

ng


Robotic

Cane

Foam

Pads

Mechani

cal

Walker None None None None None None


Distributed-Hybrid (Dh)


Alt. 1 (A1)

Alt. 2 (A2)

Alt. 3 (A3)

Alt. 4 (A4)

Alt. 5 (A5) Dh1 Dh2 Dh3 Dh4 Dh5

D1 Global

Sensing Mode

Wearab

le

Ambien

t

Hybrid

Sens

Hybrid

Sens

Hybrid

Sens

Hybrid

Sens

Wearab

le

Ambien

t

D2 Global


Wearab

le Inj

Non-

Wearab

le None

Wearab

le Inj None

Wearab

le Inj

Non-

Wearab

le

Wearab

le Inj

D3 Global

Computing Mode

Onboar

d

Offboar

d

Hybrid

Comp None

Onboar

d

Onboar

d

Hybrid

Comp

Hybrid

Comp

Onboar

d

D4 Global

Number of devices Single

Multipl

e

Multipl

e

Multipl

e

Multipl

e

Multipl

e

Multipl

e

D5 Global

Communication Mode

Short-

Range

Long-

Range

Hybrid

Comms None

Hybrid

Comms

Short-

Range

Hybrid

Comms

Hybrid

Comms

Short-

Range


Control

Center

Hybrid

Alert None

Hybrid

Alert

Hybrid

Alert

Hybrid

Alert

Hybrid

Alert

Hybrid

Alert

D7

Sensin

g

Wearable Sensing

Acceler

ometer

Gyrosco

pe +

Acc.

Baromet

er + Acc. None

Barome

ter +

Acc.

Barome

ter +

Acc.

Barome

ter +

Acc.

Barome

ter +

Acc. None

D8

Sensin

g

Ambient Sensing

mmWa

ve

RADAR UWB Camera IR

Non

e

mmWa

ve

RADAR

mmWa

ve

RADAR

mmWa

ve

RADAR None

mmWa

ve

RADAR

D9

Poweri

ng Storage Lithium

Battery

Superca

pacitor None

Lithium

Battery

Lithium

Battery

Lithium

Battery

Superca

pacitor

Lithium

Battery

D10

Poweri

ng Recharging TEG

Wireles

s

Chargin

g Wall

Kineti

c

Non

e

Wireles

s

Chargin

g

Wireles

s

Chargin

g

Wireles

s

Chargin

g TEG

Wireles

s

Chargin

g

D11

Comm

unicati

ng


Non

e BLE BLE BLE BLE BLE

D12

Comm

unicati

ng

Long-Range LTE-M LoRA SigFox 4G

Non

e LTE-M None LTE-M LTE-M None

D13

Compu

ting Onboard Tensor

Chip

Mobile

chip FPGA ASIC

Non

e ASIC ASIC ASIC ASIC

Tensor

Chip

D14

Compu

ting Offboard Cloud Edge None None None Edge Cloud None

D15

Compu

ting Algorithms Thresh

old

ML -

kNN/SV

M

Deep

Learning None

ML -

kNN/SV

M

ML -

kNN/SV

M

ML -

kNN/SV

M

ML -

kNN/SV

M

Deep

Learnin

g

D16

Interac

ting Input Buttons

Touchsc

reen Mic

Smar

tpho

ne

Non

e

Smartp

hone

Smartp

hone

Smartp

hone

Smartp

hone

Smartp

hone

D17

Interac


Smar

tpho

ne

Non

e

Smartp

hone

Smartp

hone

Smartp

hone

Smartp

hone

Smartp

hone

D18

Actuati

ng


Exoskel

eton Exosuit CMG

Non

e Airbag None Exosuit None Exosuit

D19

Actuati

ng


Robotic

Cane

Foam

Pads

Mechani

cal

Walker None None None None

Robotic

Cane None

a systems analysis and technology roadmap for fall ...systemarchitect.mit.edu/docs/reddy20a.pdfmom,...

Documents