Driving simulation and Scenario Factory for Automated Vehicle validation

Transcription

1 Driving simulation and Scenario Factory for Automated Vehicle validation Pr. Andras Kemeny Scientific Director, A. V. Simulation Expert Leader, Renault

2 INDEX 1. Introduction of autonomous driving 2. Validation of autonomous vehicle (AV) 3. Scenario factory for AV testing 4. Simulation, proving ground and field testing 5. Collaborative and public investments

3 SECTION 1 Introduction of autonomous driving on the automotive market Motivations Human vs. autonomous driving How safe is safe enough Level of introduction on the market

4 Autonomous vehicle - definition 1. Unconnected or cooperative 2. Driverless or w/ driver 3. Self-driving or under control 4. Automated vs. human operation Motivations Target of automated vehicles 1. Reduction of accidents 2. Reduction of gasoil consumption 3. Fluid traffic and Higher user rate of vehicles 4. Releasing of driver time and business opportunities 5. More space and less congestion in the cities in fine 6. New potential market opportunities

5 Motivation (suite): History of Automated Driving 1939 General Motors Futurama exhibit 1964 GM Firebird IV Futurama II exhibit 1964 Research by Fenton at OSU 1986 ROMETHEUS and PATH programs 1994 PROMETHEUS demonstration in Paris NAHS Consortium Demonstration in San Diego 2003 PATH automated bus and truck demonstrations DARPA Urban Challenge Curtesy: Steve Shladover, ITS, Berkeley

6 Human vs. automated driving fatalities Human errors as a source of 90 % of automotive crashes but The safest drivers drive 10 x better than average (age, experience, fatigue, alcohol, external causes)

7 How safe is safe enough Perception while exposition to involuntary risk: bias of risk perception (ex. flying vs. driving) Users requiring 1000 x smaller acceptable risk level The media role in risk perception (ex. Hits in Tesla accident vs. ordinary road fatalities)

8 Level of introduction on the market Regulatory mandate ex. Seat belt in the US from 40 to 90 % of adoption - New car fitting 6 years - Fitted all occupants 22 years - All occupants wearing not yet Perception of control Technological maturity with complex multiple systems with combined effects and robustness requirements

9 SECTION 2 Validation of autonomous vehicles Acceptation and deployment of autonomous vehicles depends on the extensive validation of user interface and safety level.

10 Critical ADAS and AD validation challenges PICTURE

11 From an assisted to an autonomous driving ADAS Features AD Level AD Simulation (Km/21days) Storage & compute cores 0,25 PB ON ON OFF OFF 3 M Km 500 M Km Transfer 1 TB/ day 79k cores 50 PB Transfer 1,2 PB/ day NEW SKILLS Model architects Data scientists AI PROCESS PROCESS New skills Based on virtual Model simulation architects and Data physical scientists validation AI ECOSYSTEM & PARTNERSHIP Suppliers + Contributors

12 Different levels of autonomous driving

13 Increase of the complexity of the vehicle Front camera Radar HD Map Lidar Around view camera

14 Cut in scenario EXAMPLE Cut-in vehicle : Speed > EGO and various speed conditions Preceding vehicle : Various speed conditions Trigger Distance between ego and cut-in vehicle according theirs speeds Following vehicle: Speed = EGO Speed EGO initial parameters: Speed = preceding vehicle speed Ground creation Number of lanes Width of the lanes Tilt Weather conditions Luminosity: Day time

15 SECTION 3 Scenario factory for AV testing Massive scenario generation and corresponding data analysis are necessary for thorough AV/AD testing and validation.

16 SCANeR Studio AVS Validation processes

17 PICTURE Scenario generation with the SCANeRStudio driving simulation software package.

18 PICTURE Scenario replay with the SCANeRStudio driving simulation software package.

19 Post-processing data framework Add new warnings to library Dev Team Warning Library Simulations Inputs Level 1 Main Insights of the SP Dashboards Simulation Plan Manager Warnings Level 2 Data mining on Reduced Information Regular Data Mining Data Analyst Indicators Level 3 Deep mining on Raw Data Data Mining Datalake Data Scientist Add requested indicators to library Dev Team Indicators Library

20 SECTION 4 Simulation, proving ground and field testing The stake of AD/AV validation requires an extensive mixed, simulation and physical testing procedure in order to cover all known and rare critical road traffic scenarios.

21 Mixed, simulation and physical testing and validation procedures

22 A complete AD validation chain to address growing technology complexity & Uses Case diversity Use Case catalog Capitalization & car SW Upgrade Massive SIMULATION Scenario Factory Massive Simulation platform ADIH (HPC (*)) + Human model Results analysis (Machine learning for Clusterisation and scenes recognition) Scenario Models (road, trafic, weather, ) NG On Field & Tracks data collection Accident DB / OEM DB Dysfunction collection Digital vehicle Driving Simulator + Driver in the loop Customer failed cases Vehicle Models (*) HPC: High Performance Computing NG : failed cases NG

23 AD DIL Driving simulator requirements for AD occupant perception AD Functional safety, Driver acceptance Limit conditions & dangerous Use Cases New simulation building ~2 000 m² in 2019 Example : from other direction of the highway crashes into Ego Vehicle direction Renault Investment + Example : Animal crossing the lanes Example : Motorbike cuts in front Ego Vehicle (close or far cut-in) Renault Optimized AD Simulateur (ROADS) - 9 DOF, 1G acceleration using 2 axes Screen, 3D & UHD

24 PICTURE High performance driving simulator for ADAS validation in various driving simulation. Validation of delegation (from manual to autonomous driving and vice and versa) as well as identification of rare worst case scenarios necessities the use of driving simulators.

25 Scenario identification (MOOVE Project) 1. Real world driving safety critical scenarios (SCS) 2. SCS occurrence statistics 3. New SCS 0. Use case Definition & Targeted scenarios 1. Data collect 2. Data transformation at common format 3. Calculation of high level parameter (Sensors independent) 4. Scenario searching and clustering Relative_Velocity_X Relative_Velocity_Y Absolute_Velocity_X Absolute_Velocity_Y Relative_Accel_longi Accel_longi.. Time_To_Collision Time_Between_Vehicles Status_Mobile Pos_X Pos_Y

26 Digital scenario library & test case generation (SVA Project) 1. Simulation platform 2. Digital Scenarios library

27 Digital scenario library implementation Network Partners Moove Driving data recordings SVA Platform Process and tools validation by MIL simulation Driving data recordings Accidentology databases Type approval Retex Library building Scenarios & Environments Description formats Implementation Cooperation with other consortiums (PEGASUS, SIP-ADUS, ENABLE ).

28 Driving simulation for autonomous vehicle validation

29 THANK YOU