Intelligent validation of automated driving
Tom Lueders
Director Testing Solutions
HELLA Aglaia Mobile Vision
< Schedule: Day 1, Monday April 3, 2017, 14:45 – 16:00 >
< Session: B05 -> Big Data, IoT, AI, Deep Learning >
Focus: Validation of the environment perception
Role of the TIER
• The TIER supplies sub-systems, including soft- and hardware to the OEM
• (Sub-) System Concept, Design, Development, Integration & Test derived from technical, non-technical, safety, security and legal requirements
• OEMs request to provide up to x00.000 km statistical evidence for ADAS and up to x.000.000 km for automated driving on sensor level
• Goal is to be confident that the system works in any situation
• TIER is responsible for the verification & validation at sub-system level. Incl. security, safety and performance
Test vehicle with reference sensory
Example testing vehicle sensor setup for automated driving
L I D A R C A M E R A R A D A R B U S
What are the challenges?
V e r i f y
“ k n o w l e d g e ” v s .
a l g o r i t h m C o r r e l a t e m u l t i
d o m a i n d a t a o f
a l l k i n d s
( s e n s o r s b u s ,
V 2 X , . . )
M a n a g e c o m p l e x
d a t a
L o o k a t t h e r i g h t
( i m p o r t a n t ) d a t a
P r i v a c y & D a t a
p r o t e c t i o n
I n t e l l i g e n t D a t a
P r o c e s s i n g ( T i m e ,
A l g o r i t h m s ,
e t c . )
In-vehicle recording data volume (example)
Source n Transfer Rate [Mbyte/s]
Transfer Rate total [Mbyte/s]
2 MP camera 4 98 392
8 MP camera 1 445 445
CAN 4 0,125 0,5
Flexray 2 1,25 2,5
RADAR 4 1 4
LIDAR 1 10 10
854
854 * 3600 *24 = 73.785.600 Mbyte/day (ca. 7.4 TB)
Perception performance validation – Process & Toolchain
Modular Multi-Channel Data
Recorder
Simulation Systems Integrating
Virtual Sensor Models
(Big) Data (No-SQL) Data
Management System
AnnoStation ® - Web centric QA/labeling
system
Intelligent Data Fusion & Virtual
Reference Environment Computing
Fully Automated
SIL/HIL processing
KPI based statistical
Analysis Tools
Enrich Data applying Meta
Data Extraction
Classification: Radar perception
Classification: Semantic 3D Segmentation
Classification: Semantic 3D Segmentation (2)
Web centric interactive annotation
Ground truth generation – QA teams world wide
Proposals generated using deep learning
Sensor modelling and scenario simulation
Outlook
Question #1: What topics will your work / projects address / solve?
Mission: Gain capabilites to validate autonomous vehilces
• Further build up reference environment by intelligent fusion / analysis from reference sensors and other available information (driving behavior, maps, V2X, V2V, ..)
• Continuous “critical” scenario identification & evaluation
• Continuous learning cycles combined with “in-field” validation (co-pilot)
• Intelligent compression of data (e.g. triggered recording)
• Further develop simulation capabilities
• Further develop validation processes / procedures
Outlook / Discussion
Question #2: What topics need to be addressed/solved in a short term and long term time window (year 2020 and 2040)?
• General: Definition of an international legal framework for automated driving (legal responsibilities, etc.)
• Further standardization of procedures, data formants & content and infrastructures
• Standardized infrastructures (e.g. mobile networks, data bases)
• Establish standardized urban testing areas
• European rules for homologation of automated vehicles
• International standards for data handling (e.g. protection)
• European validation standards for automated driving
Society:
• Re-think urban transportation approaches
• Re-think mobility requirements (e.g. daily driving to work? )
• Intelligent logistics (e.g. Railway vs. truck chains?)
• What's about environment protection and shrinking resources? Still valid?
TOM LUEDERS
Hella Aglaia Mobile Vision GmbH
Treskowstr. 14 13089 Berlin/Germany
Phone: +49 (0) 30 200 042 9 - 0 Fax +49 (0) 30 200 042 9 - 109 E-mail: [email protected] www.aglaia-gmbh.de
Thank you very much!