SAFETY : the biggest challenge for autonomous vehicle SIMULATION will complete actual data. Rémi BASTIEN VP prospective B RÉMI BASTIEN 2018, JUNE 1 Confidential C
AUTONOMOUS VEHICLES , WHAT STAKES? The he ke key is in n affordable transp sport modes for or us users, and nd with th hi high throughputs to o sta tay economical in n m² of of infr nfrastructure B RÉMI BASTIEN 2018, JUNE 2 Confidential C
MORE AND MORE, INFRASTRUCTURE BECOMES A KEY ENABLER Michigan initiative CAVnue B RÉMI BASTIEN 2018, JUNE 3 Confidential C
CLASSIFICATION BY LEVEL (SAE) IS UNCOMPLETE… Driver Driver Driver must Driver does not need to Driver is not System monitor the dynamic continuously continuously monitor the required during performs the driving task nor the performs the performs the dynamic driving defined use lateral and driving environment at longitudinal and longitudinal or task and the all times; however he case longitudinal must be attentive to and lateral dynamic lateral dynamic driving dynamic driving Automation ➔ Driver follow system’s requests driving task driving task environment at task in all / warnings to resume the System all times dynamic driving task. situations performs the encountered lateral and during the entire longitudinal System performs journey. No dynamic driving longitudinal and driver required. System task in all lateral driving task in performs situations in a a defined use case. longitudinal and defined use Recognizes its The other driving lateral driving performance limits case. task is task in a defined and requests driver performed by use case to resume the the system dynamic driving task No intervening with sufficient time vehicle system margin. active Level 0 Level 1 Level 2 Level 3 Level 4 Level 5 High Full Assisted Partial Conditional Driver Only Automation Automation Automation Automation *terms acc. to SAE J3016 B RÉMI BASTIEN 2018, JUNE 5 Confidential C
… AS ODD* STATUS ARE NECESSARY : EXAMPLE OF L4 𝐺𝑃𝑊 visibility = 𝑔 𝑊 Complex urban traffic (Paris Chaotic traffic Narrow FOV Warehouses Residential pathways intramuros, Rome, Madrid, (Mumbai, place de l’Etoile , etc.) etc.) ← City slow Dedicated bus Limited FOV Arterial / sub-arterial Dense intra-urban arteries War zones lanes driveways (Paris, Rome, Madrid, etc.) ← City fast Dedicated highway Non highway intercity Open FOV Intercity highways Mars surface lanes driveways (A6, A13) ← Expressways Traffic d ensity = 𝑔 𝛼 Ԧ 𝑟 No life zone Low Average High *ODD : Operational design domain B RÉMI BASTIEN 2018, JUNE 6 Confidential C
TECHNOLOGY IS MANDATORY BUT … IS NOT ENOUGH ▪ Core technologies ▪ Platform enhancement B RÉMI BASTIEN 2018, JUNE 7 Confidential C
KEY SUCCESS CONDITIONS : SOCIAL ACCEPTANCE Experimentation Social acceptance ▪ Regulations ▪ Proof by FOT on certified roads ▪ Product Liability ▪ Infrastructure ▪ Insurance ▪ Consumer awareness ▪ Driver Education ➔ THE BIGGEST STAKE IS SAFETY B RÉMI BASTIEN 2018, JUNE 8 Confidential C
AD IS FAR BEYOND ADAS, WITH AN UNCERTAIN MARKET Continuously active EMERGENCY / SAFETY Without driver intervention ASSISTANCE ACC STRESS FREE Only active under ABS / FREE TIME : LKA driver’s request ESP Park AUTONOMOUS AEB DRIVING assist Active only in limited conditions and vehicle proposal B RÉMI BASTIEN 2018, JUNE 9 Confidential C
AD IS A MAJOR DISRUPTION ADAS AD (L1, L2, L3) (L3+ (1) , L4, L5) (1) Emerging German L3 standard (Audi, BMW, Daimler) (2) Emerging consensus among European OEM Driver is the last resort System is the last resort Driver reliability proof System reliability proof Massive Driver training + experience mile accumulation + resimulation SAFETY IS A MUST B RÉMI BASTIEN 2018, JUNE 10 Confidential C
SAFETY NEEDS STANDARDS FOR DEVELOPMENT AND VALIDATION ISO 26262 defines how to assess a risk and the necessary activities to perform for each step: ❖ System ❖ Software ❖ Hardware ❖ Production... Redundancy for Autonomous Driving: ▪ Additional Safety Stakes: ❖ Redundant Sensors & Actuators ❖ For Autonomous Driving, Automotive EE ❖ Redundant Communication Networks Architecture has to switch from Fail Safe design to Fail Operational. ❖ Redundant Power supply Networks ❖ Safety has also to consider SOTIF (Safety of the Intended Functionality) B RÉMI BASTIEN 2018, JUNE 13 A Confidential C
SOTIF IS MANDATORY AND COMPLETE ISO 26262 Does a camera can identify a target in a very large roundabout without lane ? Does a radar will be accurate on a metallic bridge ? Does an ultrasound sensor can detect a child with a wool sweater? ISO 26262 Standard is necessary but not sufficient ! B RÉMI BASTIEN 2018, JUNE 14 E Confidential C
FIRST CONDITION : E/E ARCHITECTURE ABLE TO ENSURE ASILD evolutive proven B RÉMI BASTIEN 2018, JUNE 15 Confidential C
SECOND CONDITION : VALIDATION STRATEGY TO PROVE ASIL D I Statistical safety threshold II Reduction : III Road sections criticity inductors Experience plan + Simulation For each road section, calculate the « criticity cube » Order of magnitude for validation : : Nb incoming lanes x Nb exits x Strong Curvature … 20 Billions of kms Targetted, iterative physical test drive Non affordable by physical test drive Numerical simulation IV Map of road sections with criticity V Clustered road tests VI Final proof of reliability Distribution of clusters is proportional to the criticity Reliable and efficient validation Each road section is ranked by its criticity ratio = ratio of the road sections criticity cube volume / average criticity <<< 20 Billions km B RÉMI BASTIEN 2018, JUNE 16 Confidential C
EXAMPLE OF RISKY SITUATIONS ON HIGHWAY 4 1 2 3 B RÉMI BASTIEN 2018, JUNE 17 Confidential C
COMBINATION OF THE 4 SITUATIONS : VERY UNLIKELY … BUT POSSIBLE ONLY SIMULATION CAN COVER SUCH CASES WHEN MILEAGE ACCUMULATION CANNOT B RÉMI BASTIEN 2018, JUNE 18 Confidential C
KEY TAKE AWAY Autonomous Driving • Safety is a must and is extremely demanding. This will lead to first applications on limited Operational Design Domain. • ADAS potential market will stay strong before AD applications Autonomous Driving and Simulation • Simulation is essential for the development and validation • Validation time and cost will be reduced by a factor > 1000 with the support of advanced simulation AD Next challenges for simulation • To model precisely the performances of the sensors • To combine compatible scenarios to cover exhaustive risky situations • To allow safer&quicker time to market applications for ADAS and AD B RÉMI BASTIEN 2018, JUNE 19 Confidential C
Recommend
More recommend
