human machine cooperation on moon or mars … eva suit can ... each meca unit will alert the crew...
TRANSCRIPT
http://www.CrewAssistant.com
Human Machine Cooperation on Moon or Mars
Mark Neerincx, Jasper Lindenberg, Nanja Smets, Tim Grant, André Bos, Leo Breebaart, Antonio Olmedo Soler, Uwe Brauer, Mikael Wolff
http://www.CrewAssistant.com
Objective & Vision
Objective: support mission goals (without injury or loss of life) by• empowering the cognitive capacities of human-machine
teams during planetary exploration missions • in order to cope autonomously with unexpected, complex and
potentially hazardous situations.
Vision: crew support that • acts in a ubiquitous computing environment • as “electronic partner”, helping the crew
– to assess the situation, – to determine a suitable course of actions to solve a problem, – to refresh skills and knowledge,– to improve resilience.
http://www.CrewAssistant.com
Illustrative space scenario
Benny and Brendaare on a rock collecting procedure.
Suddenly,Benny’s
spacesuit fails
A Rover from another team comes to pick up Benny
Benny is brought to the habitat for recovery
http://www.CrewAssistant.com
Situated Cognitive Engineering
Prototype
HitL-test
UX
Simulation
Sim-Assess
Sim Results
OperationalDemands
Human FactorsKnowledge
EnvisionedTechnologyDerive
Refine
Review
Comments
Test
Refine
Requirements
Use Cases Claims Core Functions
Specify
contextualizeorganize justify
http://www.CrewAssistant.com
Situated Cognitive Engineering
Prototype
HitL-test
UX
Simulation
Sim-Assess
Sim Results
OperationalDemands
Human FactorsKnowledge
EnvisionedTechnologyDerive
Refine
Review
Comments
Test
Refine
Requirements
Use Cases Claims Core Functions
Specify
contextualizeorganize justify
http://www.CrewAssistant.com
Operations: Abstraction Decomposition Space (ADS) Claire Baker
Outpost Spaces/ Systems
Subspaces/ Subsystems
Components
Mission Priorties
Crew&Vehicle Safety
Values & Prio Measures
Minimise crew system maint…
Prio-related Functions
Surface Exploration
Maintenance & repair
Information dissemination
communication
Physical Capability
Base cap. to support n astronauts…
Prod. levels of ISRU plant
EVA suit can carry…
EVA suit can maintain temp. at x for y hours
Physical Objects
Mars surface outpost
ISRU plant EVA suits EVA suit thermal control…
=> decomposition
abst
ract
ion
<=
http://www.CrewAssistant.com
Human Factors
• Cognitive Task Load • Situation Awareness• Sense Making• Decision Making• Dynamic Human Capacities• Trust and Emotion• Collaboration• Crew Resource
Management
http://www.CrewAssistant.com
Technology
• Smart Task Environment with automatic distribution of data, knowledge, software and reference documents (can cope with infrastructure failures)
• Agent, Web technology and smart sensors• Model-based reasoning and health management• Human-machine (e.g. robot) collaboration and mixed
reality• Boundaries of the technology:
– maturity, – graceful degradation, – maintainability and fault tolerance, and – constraints of the hostile environment.
http://www.CrewAssistant.com
ePartner Concept that integrates operations, human factors and technology
Has information of its hPartner, e.g.• permanent characteristics (e.g., personality) • dynamic characteristics (e.g., experience) • base-line state (e.g., “normal” heart rate) • momentary state (e.g., current heart rate)• tasks to do (e.g., alarm handling)• task performance (e.g. time)• current context (e.g., location)
And interprets this information to• assess human’s condition for current context• identify critical situations (e.g. panic)• apply mitigation strategies to reduce the negative
effects (e.g. reschedule tasks, notify colleague, …)
http://www.CrewAssistant.com
ePartner’s knowledge
• Easy to share with its hPartner• Trustworthy• Based on situated cognitive theories:
– cognitive task load– emotional state– fitness– team involvement
• Continuously updating the models via human input, and automatic sensing of human behavior, physiology and context
http://www.CrewAssistant.com
Situated Cognitive Engineering
Prototype
HitL-test
UX
Simulation
Sim-Assess
Sim Results
OperationalDemands
Human FactorsKnowledge
EnvisionedTechnologyDerive
Refine
Review
Comments
Test
Refine
Requirements
Use Cases Claims Core Functions
Specify
contextualizeorganize justify
http://www.CrewAssistant.com
Theory-driven requirements engineering
requirementsrequirements
use casesuse cases claimsclaims
justifycontextualizeorganize
Situated Cognitive Theory
Requirements
Use Cases Claims Core Functions
http://www.CrewAssistant.com
ePartner Requirement Specification
Identify critical states and start mitigation:• Dialogue Style• Feedback• Crew Notification• Information Filter• Task Allocation• Automation Level• …
Example specification (simplified)
RF2024 MECA shall communicate with the crew about important events.
Claim C064 Each MECA unit will alert the crew member regarding for instance scheduled events, low-frequency nominal events and off-nominal events.+ Helps maintain high situation
awareness for crew members. Consistent notifications help maintain sufficient trust.
- May interrupt with current activities, increasing cognitive task load, and hence decreasing effectiveness or efficiency.
Use-Cases UC077, UC078, UC080, UC083
ePartner concept
http://www.CrewAssistant.com
Justification and Refinement
Claims should provide an adequate justification: • Truthful:
– information is factual – upsides, downsides & trade-offs occur as in reality – new facts => requirement is modified or removed
• Exclusive: – explains why current (not another) requirement is optimal. – alternative requirement with same upsides, downsides and
trade-offs => generalization until further research reveals which of its instantiations is the best candidate.
=> Refinement of requirements and the corresponding claims iteratively proceeds from general to specific, carefully justifying at each step the refinement made.
http://www.CrewAssistant.com
Level of Specification and Assessment
Task level• based on users’ goals and information needs, the
system’s functions and information provision are specified or assessed
• functional requirements with corresponding claims
Communication level • the control of the functions and the presentation of
the information are specified or assessed (i.e. the “look-and-feel”)
• User Interface features with corresponding claims
http://www.CrewAssistant.com
Empirical-driven refinement
The requirements derivation process:• iterative• incremental • top-down
Task level: • Adaptive notification => increased
awareness and minimal interruption
Communication level:• avatar facial expressions => valence effect• avatar speech expressions => arousal effect
“Your space suit shows a malfunction concerning the
temperature regulation. I informed Brenda and
Herman…”
E.g., adaptive notification of an ePartner
http://www.CrewAssistant.com
Use-Case-based Reqs Structuring
UC083 Alarm handling
Goal To bundle low-level alarms into meaningful events.
Actors Astronaut and MECA in habitat.
…
Reqs RF2024, RF2050, RF2080,…
Sequen ce
1. Low-level smoke and IR alarms. 2. MECA combines the low-level alarms to determine location/scope of fire. 3. MECA gets attention of astronaut 4. MECA provides procedures 5. Astronaut fights fire 6. Fire is put out.
Example use-case (simplified).
A
B
C
D
EF
G
H
A B C D E HGF
(b)
(a)
Clustering of 8 reqs
hierarchical clustering: similarity based on use-cases to which the reqs apply
hierarchical clustering: similarity based on use-cases to which the reqs apply
http://www.CrewAssistant.com
Use-Case-based clustering
Alternative or complement for functional decomposition• really user-centered• dynamic and flexible• can highlight functional-usage structure mismatches
Optimizing evaluation for testability and empirical value:• The more requirements in a hypothesis, the more closely it
resembles the design specification => the higher its empirical value, but the lower its testability.
• Related requirements have interdependencies and apply to similar situations => increases its empirical value and testability
Selecting hypotheses for empirical evaluation:• Select an “under-evaluated” hypothesis• If the hypothesis is very difficult to test, move down • If the hypothesis is very easy to test, move up
http://www.CrewAssistant.com
Dendogram
Nodes represent clusters/hypotheses• Red: not plausible yet • Green: plausible
Nodes represent clusters/hypotheses• Red: not plausible yet • Green: plausible
http://www.CrewAssistant.com
Situated Cognitive Engineering
Prototype
HitL-test
UX
Simulation
Sim-Assess
Sim Results
OperationalDemands
Human FactorsKnowledge
EnvisionedTechnologyDerive
Refine
Review
Comments
Test
Refine
Requirements
Use Cases Claims Core Functions
Specify
contextualizeorganize justify
http://www.CrewAssistant.com
Evaluation
Fidelity • adequate representation of relevant rules, particularly
dependencies, in a human-agent team
Realism • varies from one extreme—the real environment—to the
other—a virtual environment.
Evaluation Options• Computer simulations• Mixed Reality lab experiments• Real life experiments in analogue environments• Real life Experiments in a 500 day isolation experiment.
http://www.CrewAssistant.com
Increasing fidelity and reality
Reality
Fide
lity
Virtual Real
Low
Hig
h
Computer simulated
Computer simulated & policies
Human-in-
VR-loop
Moon/Mars
WOz
Analogue environment
http://www.CrewAssistant.com
Ontologies
Policies
agent 1agent 2agent 3
… KAOS: Simulation of Collaboration.• Obligations, authorizationsBrahms: Simulation of Work Practice.• Agents, …
Computer Simulations
http://www.CrewAssistant.com
Evaluations in analogue environments/ isolation
MARS500: gaming as diagnostic and training tool for resilienceAnalogue environments (e.g.,
Eifel volcanic area): ePartner helps to cope with (social, cognitive and affective) demanding situations.
Concordia: maintaining physical and mental fitness
DoneDone
In ProgressIn Progress
PlanPlan
http://www.CrewAssistant.com
MARS-500
• Collaborative Trainer (COLT)– Multi-user – 1 instructor and 2 trainees– Learn to use systems (water tank,
cardiopres) with a procedure viewer
• Colored Trails– Multi-user – Proposers and responders – Position pieces onto or as close as
possible to a goal location
• Lunar Lander– Single-user– Land a Lunar Lander on the Moon
A M H AAutomatic Mental Health Assistance
http://www.CrewAssistant.com
Cognitive Task Load
Level of Information Processing:How complex is the task for the actor?• Low: routine (“automatic”)• High: non-routine (“intensive
problem-solving”)
Task-Set Switches:How often must the actor switch?• Low: one task after the other• High: continuous interleaving or
interruption
Time occupied:How much time for small pauses?• Low: slowing down/ pauses are
allowed• High: must perform at maximum
speed to complete the tasks
http://www.CrewAssistant.com
Emotion: Valence and Arousal
Arousal
Valence
high
positivenegative
low
Relaxed/soothing
Delight/rejoice
gloomy
Terrified/restless
Arousal (exciting/calm)Valence (displeasure/pleasure)
http://www.CrewAssistant.com
Modeling Cognitive Task Load and Emotion (e.g. with Bayesian Networks)
LIP
U/Oload
Events …
Cognitive Task Load
Performance
Emotional State
TO
Tasks
Happiness
Speech…
Face
Valence ArousalTSS
CogLockUp…
…Anger
observables
classe
s
dimensions
Identify critical states:• intra & inter models
http://www.CrewAssistant.com
Bayesian Networks
• Analysis existing data sets– Process control tasks in laboratory
• 12 students: platform supervision, damage control & navigation
• 1407 cases (60 sec with an overlap of 50 sec)• Performance prediction: 84.8%
– Process control task on sailing ship• 24 participants: 3 scenario’s on sailing ships• 3478 cases (60 sec with an overlap of 50 sec)• Performance prediction: 74.2%
• MECA m500 software and protocol– 24 participants performed 306 cases– Performance prediction:
• CTL: 59.2%• ES: 57.8%• CTL + ES: 67.0%
http://www.CrewAssistant.com
Conclusions
• The MECA project is providing a Requirements Baseline (RB) with prototypes for enhancing human-machine resilience– Several refinements have been established
• The situated Cognitive Engineering Method helps to establish a theoretical and empirical sound RB– Tools are needed for updating and maintaining the RB
• Experiments with prototypes are providing – Performance data and user feedback on the ePartner functions– Data to train the ePartner models on Cognitive Task Load, Emotion,
Social Networks and Performance.
• MECA provided a software platform that combines simulation and Human-in-the-Loop evaluation in a flexible way– Combination with Virtual Reality to be done
• Scientific progress is being shown in numerous publications
http://www.CrewAssistant.com
Next Steps
• Requirements baseline tools for– sharing (e.g. for implementation)– maintenance (e.g. for version management)– refinement (e.g. for traceability)– organization (e.g. for evaluation guidance)
• Experiments– The 520-days study in MARS-500– Concordia– International Space Station
• Prototypes– Automatic state sensing– User modeling– Adhering to policies for human-machine collaboration– Persuasion of desirable behaviour– Supporting of refreshment training