a n d a t a effectiveness traffic efficiency driving slow with low density driving faster with high...
TRANSCRIPT
A N D A T AA N D A T A 1
Dr. Andreas Kuhn
A N D A T A
Scenario-based Validation of Automated Driving Functions
with MATLAB
Stuttgart, 2019-04-11
A N D A T AA N D A T A 22
Fields of Competence
• Artificial Intelligence
• Data Mining
• Big Data Analytics
• Modeling and simulation
• Predictive Model based
Control
• Distributed Control
• Signal Classification
• Swarm Intelligence
• (Embedded) Software
• Decision Support Systems
• Robustness and Complexity
Management
Contact: A-5400 Hallein, Hallburgstraße 5, +43 6245 74063, [email protected], www.andata.at
A-1050 Vienna, Straussengasse 16
Data driven
development
process
Automotive safety
(Mobile) Robotics Automated Guided
VehicleIndustry 4.0,
Digitalization
Failure prediction
Big Data Analytics &
AI & Simulation
Anomalies and
Incident Detection
Vehicle and traffic
automation
since 2004
© 2019, ANDATA, several granted and pending patents
A N D A T AA N D A T A 3
➢ changing traffic modes
➢ choosing the right traffic mode
according
the given
situation
Expected Benefit Categories for Automated Driving
Benefits of
Automated
Driving
Comfort
Safety
Traffic
Effectiveness
Traffic
Efficiency➢ driving slow
➢ with low density ➢ driving faster
➢ with high density
➢ more responsive
The only
intrinsic
feature!!!
➢ e.g.
automation
levels 3
Vehicle
Efficiency
➢ less responsive
➢ driving slow
➢ high density for
wind shadow
➢ Validation needs quantitative description of the pareto fronts!
© 2019, ANDATA, several granted and pending patents
A N D A T AA N D A T A 4
Main Entities Effecting the Performance of Automated Driving
© 2019, ANDATA, several granted and pending patents
Traffic Control &
Management Systems
ADAS &
Automated Driving
Traffic Situation
Behavior of Drivers &
Traffic ParticipantsLaws & Guidelines
InfrastructureQuantification of
Dependencies
A N D A T AA N D A T A 5
Main Entities Effecting the Performance of Automated Driving
© 2019, ANDATA, several granted and pending patents
Traffic Control &
Management Systems
ADAS & Automated
Driving
Traffic Situation
Behavior of Drivers &
Traffic ParticipantsLaws & Guidelines
Infrastructure
Control
Algorithm
Control
Algorithm
Co
op
era
tion
A N D A T AA N D A T A 6
Methods of Choice in Development and Validation of
Automated Driving Functions
© 2019, ANDATA, several granted and pending patents
Technology/Method
Integral and Holistic Top-Down Development
Procedures
Targets driven KPIs
Scenario- & Simulation-based Development
Approach
Data driven Simulations
Machine Learning & AI Referable
performance
Control
Functions
Prospective Effectiveness Assessment Effective Ratings
A N D A T AA N D A T A 7
Connecting Austria
• Lead Project for Automated Driving in Austria
• Platooning as instrument for improved energy and traffic efficiency
• Development and assessment of cooperative, connected, (semi-)
automated driving strategies
• 4 Principal Scenarios
© 2019, ANDATA, several granted and pending patents
© S
warc
oF
utu
rit
A N D A T AA N D A T A 8
R&D Approach / Procedures
Comfort SafetyTraffic
Effectiveness
Traffic
Efficiency
Vehicle
Efficiency
Theoretical Potentials and Effectiveness
Practical Potentials and Effectiveness
Risks and Potential Counter Effects
Total Multidisciplinary EffectivenessTra
ffic
Ob
se
rva
tio
n
Natu
ralis
tic
Drivin
g
Laws &
Guidelines
Vehicle Control
Strategies
Infrastructure & V2X &
Traffic Control
Dynamic Road
Risk Map
Sce
na
rio
Ma
na
ge
ment
© 2019, ANDATA, several granted and pending patents
A N D A T AA N D A T A 9
Scenario-Management and Development/Approval of Actions
© 2019, ANDATA, several granted and pending patents
A N D A T AA N D A T A 10
Va
r. L
ab
elli
ng
s
Variation of Systems/ComponentsVariation of Situations / Conditions
Variation of ActionsVariation of Behaviours
VERONET Solution Concept for Automated Driving and Traffic
© 2019, ANDATA, several granted and pending patents
Scenario Catalog
Simulations
Naturalistic Driving
Connected Mobility
Test Fields
Ve
hic
les
Tra
ffic
Art
ific
ialIn
telli
ge
nce
Big
Da
ta A
naly
tics
Ro
bu
stn
ess
Mn
gt
Co
mp
lexity
Mn
gt
Eff
ective
ne
ss
Ra
ting
Re
fere
nce
Syste
ms
Syste
m-o
f-S
yste
ms
Co
nflic
tA
naly
sis
Co
nd. P
roba
bili
stics
➢ Quick Identification and Resolution of
Requirement Conflicts
➢ System Understanding
➢ Conform Specification of Components
➢ Realistic Performance Ratings
➢ Excellent Control Algorithms
The Data Cube(s)
for Vehicle and
Traffic Automation
10
A N D A T AA N D A T A 11
Folding Various Decision Variables (e.g. collision probabilities)
© 2019, ANDATA, several granted and pending patents
11
A N D A T AA N D A T A 12
Scenario Management for Traffic Automation
• Development/approval of cooperative vehicle and traffic control
• Building up a scenario database including variations of
• traffic situations (volume and composition)
• traffic control algorithms
• vehicle control algorithms
• interactive behaviours of drivers,
pedestrians, etc.
• road conditions, weather
• road/lane geometries/topologies
© 2019, ANDATA, several granted and pending patents
A N D A T AA N D A T A 13
Identification and Clustering of the Relevant Traffic Situations
© 2019, ANDATA, several granted and pending patents
Map data © 2017 Google
Automated evaluation of floating car data
and various traffic sensors
A N D A T AA N D A T A 14
Adaption
Operation
Development/Training
Procedure for „Connected Development“
© 2017, ANDATA, confidential, several granted and pending patents
14
x1
x2
x3
…
xn
x y
Sensors &
InformationAction
y1
y2
y3
…
yn
xn+1x y
yn+1
Anomalie-
Detection
Offlin
eO
nlin
e
Serv
er/
Offic
eC
lient/
Onsite
Backend
Fro
nte
nd
➢ Collective (Co-)Learning Systems
Function
Algo
Function
Algo
yn+1
Robustn
ess
Mana
gem
ent
A N D A T AA N D A T A 15
Data Acquisitions from Fleet Data
© 2017, ANDATA, several granted and pending patents
15
A N D A T AA N D A T A 16
Evaluation of Tracks from Single Vehicles
➢ Development of individual,
context aware behavioural
driver models
➢ Spatial statistics of street
sections
➢ Validation of
microsimulation model
© 2019, ANDATA, several granted and pending patents
A N D A T AA N D A T A 17
Evaluation of Tracks from Single Vehicles
➢ Development of individual,
context aware behavioural
driver models
➢ Spatial statistics of street
sections
➢ Validation of
microsimulation model
© 2019, ANDATA, several granted and pending patents
A N D A T AA N D A T A 18
Traffic Observation with Video Tracking and Further Sensors
• Generating tracks of all
traffic participants by video
• Fusion with further
sensors (radar, lidar, ToF)
in work
➢ Data for development of
interactive behavioural
models (extrinsic)
© 2019, ANDATA, several granted and pending patents
© SCCH © SCCH
VRU tracks Vehicle tracks
A N D A T AA N D A T A 19
Comparision of different control variants and their effects
wrt traffic performance for a selected traffic situations
• Statistics of
Key Performance Indices
• including robustness
assessment
© 2019, ANDATA, several granted and pending patents
A N D A T AA N D A T A 20
Comparision of different control variants and their effects
wrt traffic performance for a selected traffic situation
Reference solution Improved solution
© 2019, ANDATA, several granted and pending patents
A N D A T AA N D A T A 21
Postprocessing and Rating of Different Control Strategies
Mainly Human Drivers Mainly Robot Drivers
© 2019, ANDATA, several granted and pending patents
A N D A T AA N D A T A 22
Dynamic Risk Rated Map
Adaptive wrt
• local conditions
• traffic situation
• weather
• temporal incidents
© 2019, ANDATA, several granted and pending patents
A N D A T AA N D A T A 23
ANDATA Software and Tools
© 2019, ANDATA, several granted and pending patents
23
• Data collection, preparation and
normalization
• Data cleaning
• Sensor models
• Signal preparation
• Requirements definition ("labelling", etc.)
• Data analysis
• Training, adaption and evaluation of
Machine Learning models
• Meta modelling, feature selection, etc.
• Scenario management
• Multilevel stochastic simulation
• Execution of distributed simulations
• Data plausibilization
• Anomalies and incident detection
A N D A T AA N D A T A 24
Summary and Conclusion
• Automated driving functions cannot be validated effectively without
massive utilization of numerical simulation
• Scenario Management and simulation-based system assessment is
core technology for functional validation
• Example/data based approaches are the most effective for complex
and multi-disciplinary functions
• Intrinsic requirement conflicts for automated driving functions cause
the must to multi-criteria optimization respectively pareto surfing
• MATLAB is a supreme platform to bring arbitrary disciplines and
domains into one common environment for development and
assessment
© 2019, ANDATA, several granted and pending patents
A N D A T AA N D A T A 25
A N D A T A GmbHDr. Andreas Kuhn
Tel: +43 6245 74063
Email: [email protected]
Web: www.andata.at
25
Thanks, for listening!
The singularity is near, let‘s be prepared!
© 2019, ANDATA, several granted and pending patents