Domain model based derivation of virtual driving scenarios for vehicle simulations from in-field measurements

Transcription

1 Domain model based derivation of virtual driving scenarios for vehicle simulations from in-field measurements Johannes Bach, FZI Research Center for Information Technology Advanced Microsystems for Automotive Applications (AMAA) 2016 September 23, 2016 Representation of the State of Baden-Württemberg to the European Union, Brussels, Belgium FZI FORSCHUNGSZENTRUM INFORMATIK

2 Agenda Motivation Concept for scenario abstraction Proposed modelling approach Prototype implementation Conclusion & future work 9/26/2016 FZI Forschungszentrum Informatik 2

3 The 10 9 mile issue How to validate automated vehicles? Real world testing Simulation based testing Test cases are always valid Highly expensive Can be dangerous Poorly reproducible Reproducibility is ensured Initially expensive Needs validation Which variations are interesting? 9/26/2016 FZI Forschungszentrum Informatik 3

4 Development of automotive systems and software Real World Driving Test Direct replay with open-loop behavior Reactive simulation environment Data logging device How to use recorded data for test and simulation? Abstraction and resimulation with closed-loop behavior and optional variations 9/26/2016 FZI Forschungszentrum Informatik 4

5 Abstraction of an ACC-Approach-and-Follow Scenario ACC-Approach-and-Follow Approaching a slower moving lead vehicle and slowing down to follow by distinct time-gap FZI Forschungszentrum Informatik 5

6 MOTIVATION DRIVING SCENARIO ABSTRACTION MODELLING APPROACH PROTOTYPE CONCLUSION & FUTURE WORK Abstraction of an ACC-Approach-and-Follow Scenario Discrete recording Temporal synchronous recording of a driving sequence represented by singular data snapshots Vehicle Velocity [m/s] Time [s] Vehicle Position[m] Time [s] FZI Forschungszentrum Informatik 6

7 MOTIVATION DRIVING SCENARIO ABSTRACTION MODELLING APPROACH PROTOTYPE CONCLUSION & FUTURE WORK Abstraction of an ACC-Approach-and-Follow Scenario Abstraction Time-based sequence of snapshots Vehicle Velocity Abstraction of[m/s] the scenario Time [s] Identification and clustering of logical cohesive sequences Vehicle Position[m] Time [s] FZI Forschungszentrum Informatik 7

8 MOTIVATION DRIVING SCENARIO ABSTRACTION MODELLING APPROACH PROTOTYPE CONCLUSION & FUTURE WORK Abstraction of an ACC-Approach-and-Follow Scenario Abstraction Time-based sequence of snapshots Vehicle Velocity [m/s] Trailing vehicle decelerates Lead vehicle is followed at constant time gap Vehicles come to a halt Act 1 Act 2 Act 3 Act 4 Free Cruise Approach Follow Follow Free Cruise Free Cruise Free Cruise Stop Both vehicles cruising with constant velocity Time [s] Vehicle Position[m] Sequence of abstract acts with non-varying driving maneuvers Time [s] Trailing vehicle Lead vehicle

9 Main entities of the domain model Geyer et al propose a movie and theatre related terminology S. Geyer, M. Baltzer, B. Franz, S. Hakuli, M. Kauer, M. Kienle, S. Meier, T. Weißgerber, K. Bengler, R. Bruder, F. Flemisch, and H. Winner, Concept and development of a unified ontology for generating test and use-case catalogues for assisted and automated vehicle guidance, IET Intelligent Transport Systems, vol. 8, no. 3, pp , Use Case TestCase TestCase Route Scenario Route Situation Scenario Scene Ego Vehicle Dynamic Elements Scenery Instructions Driver Action Autom. Action Driving Mission FZI Forschungszentrum Informatik 9

10 Main entities of the domain model Scenario Postulated development of events Encompasses a static scenery Act Grouping of similar snapshots Temporal abstraction by sequencing Participants Spatial abstraction by perception and interaction layers Maneuver Behavior abstraction by maneuver types One maneuver transition per act Events Specify the exact maneuver transition All events are based on relative relations to other participants or the state at the start of the act near field mid field far field free cruise approach following stop Spatial abstraction FZI Forschungszentrum Informatik 14 Temporal abstraction act 1 act 2 act 3 act 4 P1 P1 P1 P1 P2 P2 P2 P2 event: dist rel < s 0 event: vel rel < dv event: time > t end

11 Modeling approach 4 layer approach based on Eclipse Ecore M3 Ecore Meta-meta-model specification of the used modelling language Abstraction M2 M1 Mapping LogDaSim Domain Model Scenario Instances Rule Graphs Meta-model Specification of the entities within the modelled domain Model instances Representation of real world instances and objects using the domain model M0 Recorded Drving Data Executabel scenarios Real objects Recordings of real test drives and runnable simulation specifications 9/26/2016 FZI Forschungszentrum Informatik 15

12 Prototype scenario editor Specification of act based scenarios Import of recorded data Export of derived simulation scenarios Modelling of maneuver types based on situations 9/26/2016 FZI Forschungszentrum Informatik 16

13 Prototype scenario editor Abstract specification of maneuver types based on logic-level situations Spatial abstract conditions by perception layers Relative relations between participant s states for situational specification Free-Cruise Empty lane in front and left of ego-vehicle Approach Preceding vehicle with V 1 < V Ego Follow Preceding vehicle with V 1 ~ V Ego 9/26/2016 FZI Forschungszentrum Informatik 17

14 Import of recorded data Mapping of communication matrix with vehicle state and perception layers LogDaSim Domain Model Hysteresis thresholds to analyse conditions within situations Mapping Scenario Instances Rule Graphs Recorded Drving Data Maneuver identification via rule-solving over situations 9/26/2016 FZI Forschungszentrum Informatik 18

15 Export of derived simulation scenarios Substantiation by Parametrization of physical event variables LogDaSim Domain Model Verification by evaluation of parameters on logical level Mapping Scenario Instances Rule Graphs Substitute variables for test variation and permutation Carmaker Testruns 9/26/2016 FZI Forschungszentrum Informatik 19

16 MOTIVATION RELATED WORK DRIVING SCENARIO ABSTRACTION DOMAIN MODEL & Conclusion and future work Real world driving data Direct usage for open-loop replay Reactive simulation environment Data logging device Driving Scenario modelling Ecore Parametrable scenarios Abstraction of recorded data Specification of scenario filters LogDaSim Domain Model Mapping Scenario Instances Rule Graphs Recorded driving data Carmaker Testruns Generation uses Tests 9/26/2016 FZI Forschungszentrum Informatik 20

17 MOTIVATION RELATED WORK DRIVING SCENARIO ABSTRACTION DOMAIN MODEL & Conclusion and future work CONCLUSION Concept for consistent spatial, temporal and behavioral abstraction Feasibility proven with approach-and-follow scenario Scenario Import from measurements possible Generation of permutable simulation scenarios NEXT STEPS Evaluations with multi-participant scenarios Definition and evaluation of further maneuver types Integration of static scenery elements as reference points 9/26/2016 FZI Forschungszentrum Informatik 21

18 The authors Dipl.-Ing. Johannes Bach Research Scientist Embedded Systems and Sensors Engineering (ESS) FZI Research Center for Information Technology Haid-und-Neu-Str D Karlsruhe Tel.: Fax: The LogDaSim project has been funded by the German Federal Ministry of Education and Research (BMBF) under the grant 01IS14024C M.Sc. Stefan Otten Deparment Manager Embedded Systems and Sensors Engineering (ESS) FZI Research Center for Information Technology Haid-und-Neu-Str D Karlsruhe Tel.: Fax: Prof. Dr.-Ing. Eric Sax Scientific Director Embedded Systems and Sensors Engineering (ESS) FZI Research Center for Information Technology Haid-und-Neu-Str D Karlsruhe Tel.: Fax: FZI Forschungszentrum Informatik 22