CONNECTED AUTOMATION HOW ABOUT SAFETY? Bastiaan Krosse EVU Symposium, Putten, 9 th of September 2016
TNO IN FIGURES Founded in 1932 Centre for Applied Scientific Research Focused on innovation for 5 societal themes 3.000+ employees world-wide Annual turnover of 600+ million Global presence and projects Connected automation 09 September 2016
TNO AUTOMOTIVE RESEARCH Supporting the automotive industry with innovative solutions, using: In-depth knowledge and expertise in multiple domains Strong portfolio of Intellectual Property and White Box Software Algorithms Highly skilled and experienced workforce Unique state-of-the-art research facilities Advanced tools & methods for fast and robust application development Focus on accelerating the deployment of cooperative automated driving systems in various application domains (car, truck, bus, AGV) Connected automation 09 September 2016
THE NEAR AND THE FULL POTENTIAL TNO s VISION: The full potential of automated driving can only be reached by combining cooperative driving (V2x enabled) and automated driving to improve: Vehicle and traffic efficiency Comfort Safety Source: Ministry of Infrastructure &Environment / RWS 2016 Connected automation 09 September 2016
AUTOMATED VS COOPERATIVE AUTOMATED Google car in desert vs C-ACC Toyota Prius on Dutch highway Action reaction Intention coordinated reaction Conclusion: Automated vehicle: priority on interacting with environment (reactive) Cooperative vehicle: priority on understanding traffic behavior, taking coordinated action (proactive) Connected automation 09 September 2016
COMMUNICATION SUPPORTS SAFETY No communication between vehicles = no understanding of each others intention Result: action results easier in over-reaction Effect: (too) late (emergency) braking accidents and/or (ghost) traffic jams bad cooperation on high ways unsafe and inefficient merging Late detection of presence of vulnerable road users unsafe urban intersections Unclear interaction with safety vehicles unsafe and inefficient for both safety and normal vehicle. and many more unsafe scenarios Cooperative automated driving systems makes these type of typical traffic situations safer (no overreaction = smoother traffic = less chance for accident) Boundary conditions: To build a safety application, the supporting systems need to be functionally safe under all conditions, in all scenarios. How do we assure this? Connected automation 09 September 2016
SOME EXAMPLES OF COOPERATIVE AUTOMATED DRIVING (CAD) Connected automation 09 September 2016
EXAMPLE I COOPERATIVE AUTOMATED CRUISE CONTROL (C-ACC) Vehicles drive at short and constant inter-vehicle distance Desired accelaration is shared Inter vehicle distance down to 0,25 sec. Immediate braking, safe, String stable = no harmonica effect i.e. no exitation = increased safety Connected automation 09 September 2016
EXAMPLE II INNOVATIONS IN CYCLIST SAFETY SOLUTIONS INFRASTRUCTURE PERSPECTIVE: Communication of information collected by the infrastructure cooperative technology Car with C-AEB Road Side Unit Road Side Unit communicates the bicycle trajectory to the approaching car through WiFi-p Car calculates risk of collision In case of risk, a warning is issued to the driver and automatic breaking is applied for the car A newly developed intelligent bike warns the cyclist with haptic signals to hazards identified by the Road Side Unit Infrastructure sensors can be used to get a detailed insight in common cyclist behaviour patterns Cyclist Connected automation 09 September 2016
EXAMPLE III COOPERATIVE AUTOMATED MERGING (GCDC 2016) Builds on GCDC 2011 platooning skills Automated vehicles Level 1 or Level 2 Negotiation required to optimize merging of 2 platoons before construction site Challenge is to negotiate and perform the merge before the lane is closed For evaluation the scenario will be repeated multiple times with different constellations of teams Connected automation 09 September 2016
EXAMPLE IV COOPERATIVE AUTOMATED GAP MAKING FOR SAFETY VEHICLES (GCDC 2016) Demonstrates a traffic situation where cooperative vehicles reduces risk and maintains traffic flow Potential to use existing cellular technology and mobile app Using geo-networking to target relevant road sections Movie http://www.gcdc.net/en/ Connected automation 09 September 2016
NICE EXAMPLES, BUT HOW DO WE ASSURE FUNCTIONAL SAFETY OF THE UNDERLYING CAD SYSTEMS UNDER ALL CONDITIONS? How do we collect all the relevant scenarios that represent all conditions? Which scenario s are most severe? How often do they appear? Streetwise approach Connected automation 09 September 2016
CURRENT DEVELOPMENTS FOR AUTOMATED DRIVING SYSTEMS (incl. ADAS), ALL SCENARIOS BECOME IMPORTANT : Traditionally, test cases are designed based on accidentology For ADS in general, also frequent every day traffic scenarios (with multiple parameters) need to be considered: turning left bicycle car going straight turning right Most relevant conflict scenarios 6 >36 Relevant interaction scenarios Many cyclists are present in the scene Multiple MIOs (most important objects): Type of MIO Speed profile Past trajectory > predicted trajectory Behaviour > focussed vs. undetermined Infrastructure related parameters (type of road, intersection, traffic rules, view-blocking obstructions, ) Presence and manoeuvres of other road users in the scenery Disturbances (weather/lighting conditions, road works, ) Connected automation 09 September 2016
#km / scenario on the road CHALLENGE Determine real-life performance without driving multiple million km: Data collection under real-life conditions on the road Data analysis: scenario identification & classification (big data solutions required) Use library of scenarios in virtual testing for development and implementation Scenario: A typical maneuver on the road with the complete set of relevant conditions and trajectories of other traffic participants that have an interaction with the host vehicle over a relevant time period (order of seconds) A ride on the road can in this way be described by a continuous sequence of scenarios Scenarios might partly overlap in time Criteria for scenario distinction / classification Criterion of required # km / scenario Connected automation 1 i Scenario identifier N 09 September 2016
HOW TO COLLECT THE RELEVANT SCENARIOS? Connected automation 09 September 2016
Reference measurements with ground truth: - vehicle - infrastructure Real life vehicle data STREETWISE ARCHITECTURE IDENTIFY, CLASSIFY AND STORE SCENARIOS Accident databases UDRIVE Euro NCAP Big Data Vehicle data Sensor data Vehicle data Sensor data Vehicle data Sensor data # RAW DATA STORAGE - vehicle - sensors - annotations Run identification and classification algorithms SCENARIO DATABASE incl. all characteristic parameters Calibration based on ground truth Develop classification algorithms (deep learning) Select scenarios Vary parameters BEHAVIOUR MODELS - pedestrians - bicyclists - vehicles Select various target behaviour EVALUATION OF (C) AD systems: Virtual testing (extensive set) Physical testing (limited # of km) Movie scenario generation: https://www.youtube.com/watch?v=zmxefcumrse Connected automation 09 September 2016
CONCLUSION Cooperative Automated Driving (CAD) will bring the full potential of vehicle and traffic safety To make CAD (functionally) safe under all relevant conditions (road, weather, traffic) we need an efficient approach to collect, identify and classify all relevant scenarios Connected automation 09 September 2016
THANK YOU FOR YOUR ATTENTION QUESTIONS? BASTIAAN.KROSSE@TNO.NL