SIP-adus Workshop 2018 A Traffic-based Method for Safety Impact Assessment of Road Vehicle Automation Tokyo, 14 th November 2018 Dr.-Ing. Adrian Zlocki, Christian Rösener, M.Sc., Univ.-Prof. Dr.-Ing. Lutz Eckstein Forschungsgesellschaft Kraftfahrwesen mbh Aachen Slide No. 1
Motivation Public View on Automated Driving How SAFE is Automated Driving? Research Question What is the safety level of automated driving? Methodology A Traffic-based Method for Safety Impact Assessment of Road Vehicle Automation Slide No. 2
Evaluation Methodology Impact Assessment vs. Safety Assurance I M P A C T A S S E S S M E N T Effectiveness Usability Acceptance Potential conflict Safety of Interaction Efficiency Controllability & Functional Safety SOTIF Behavioral Safety Operational Safety S A F E T Y A S S U R A N C E Slide No. 3
Analysis of Automated Driving Field Test Data Scenario Classification of Real-World Data Variation of measured situations New generated situations Parameter x Parameter x Parameter z Parameter y Accident data FOTs Variation of measured situations + Generation of new situations Source: Eckstein, L., Zlocki, A.: Safety Potential of ADAS - Combined Methods for an Effective Evaluation, 23rd ESV 2013, Seoul, 2013 Slide No. 4
Impact Assessment of Automated Driving Driving Scenarios from Accident Type Example: Passive Cut-In 646 - Overtaking 643 Decelerating Cut-In (left) Overtaking on carriageway 631 Decelerating Cut-In (right) Slide No. 5
Impact Assessment of Automated Driving Driving Scenarios from Accident Type Approach: The types of driving scenarios, respectively physical accident constellations, do not change with automated driving. The frequency of occurrence and the severity of these driving scenarios may change with automated driving. Slide No. 6
Impact Assessment of Automated Driving Definition of Methodology for Impact Assessment 3 parameter i Driving scenario-based estimation of effectiveness field S 1 S n 1 2 Automated driving function Definition of Scenarios 4 parameter n Ref AD f(s n ) S 2 Frequency of occurence of driving scenarios 3 Frequency of Occurrence Scenarios Derivation of Effectiveness Field Severity in Scenarios 4 5 6 Impact of automated driving function 5 parameter i Severity in driving scenario parameter n Slide No. 7
200 m 20 20 10 m 10 m Fahrtrichtung 60 80 m 20 200 m Impact Assessment of Automated Driving 1 Definition of automated driving function and 2 scenarios 1 2 Motorway-Chauffeur Automation level (SAE): 3 Operational design domain (ODD): Operation domain: Relevant driving scenarios: Driving accident Approaching static Object Approaching vehicle Traffic Jam Lane change Cut-in Take over Approaching lateral Object Crossing intersection Turn around Turn at intersection Slide No. 8
Impact Assessment of Automated Driving 3 Effectiveness Field and Scenario Classification Scenario Approaching vehicle Parameter x Parameter z v Accident data FOTs Scenario Cut-in of other vehicle Scenario Lane change Slide No. 9
Driving scenario-based estimation of effective. field 3 Accidents in Germany according to ODD Accidents in Germany in 2016 308.145 A(P) Accidents in domain Motorway 19.010 A(P) Rural road Municipal road 38% Effectiveness field 47% County road 14% Federal highway 17% Federal motorway 6% not adressable driving scenarios 14% driver and vehicle related limits 11% functional limits 17% no car participation 11% Slide No. 10
Driving scenario-based estimation of effective. field 3 Input data for scaling-up and simulation Approaching leading vehicle 43 % Driving scenarios in effectiveness field 9.395 A(P) Approaching static object 2 % Lane change 4 % Approaching traffic jam 18 % Driving accident 17 % Accident statistics GIDAS FOT Number of accidents for scaling-up of effectiveness Driving scenario Accidents Passive cut-in 1.219 Situational variables for effect of function by simulation v 1 =67 km/h Passive cut-in 16 % v 2 = 93 km/h Slide No. 11
Safety Impact Assessment of Automated Driving 4 Identification of frequencies of driving scenarios Automated driving function Frequencies of driving scenarios Ref Definition of driving scenarios Driving scenario based estimation of effectiveness field AD f(s n ) Frequency of driving scenario Severity in driving scenario Impact of automated driving function Slide No. 12
Identification of frequencies of driving scenarios 4 FOT-data AD ika Ref AD Driving scenario classification frequencies of scenarios Slide No. 13
Impact Assessment of Automated Driving 4 Identification of Frequency from FOT Data AD Slide No. 14
Identification of frequencies of driving scenarios Traffic simulation 4 Slide No. 15
Severity in driving scenarios by re-simulation 5 Simulation framework Human driver performance models from driving simulator study/fot for reference Driving scenario passive cut-in Ref AD Simulation of reference Simulation with ADF Severity Slide No. 16
Safety Impact Assessment of Automated Driving Impact Assessment Results 6 90% Effectiveness in domain 80% 70% 60% 40% 30% 20% 17% 30% 53% 13% 24% 46% 14% 26% 14% 26% 54% 10% 0% 1% 3% 8% 4% 4% 4% 3% Traffic Jam- Chauffeur Motorway- Chauffeur Commuter Chauffeur Universal- Chauffeur Urban Robot- Taxi Slide No. 17
Safety Impact Assessment of Automated Driving Impact Assessment Results 6 300,000 Accidents with personal injuries per year 250,000 200,000 150,000 100,000 50,000 0 accidents in domain, of them without involvement of passenger car outside the functional limits none or ambiguous effect avoided Traffic Jam- Chauffeur Motorway- Chauffeur Commuter Chauffeur Universal- Chauffeur Urban Robot- Taxi Slide No. 18
Safety Impact Assessment of Automated Driving 6 Impact Assessment Results Accidents with personal injuries per year 300.000 250.000 200.000 150.000 100.000 50.000 0 accidents in domain, of them without involvement of passenger car outside the functional limits none or ambiguous effect avoided X % effectiveness in domain 1% 3% 8% 4% 17% 30% 53% 4% 13% 24% 46% 4% 14% 26% 3% 14% 26% 54% Traffic Jam- Chauffeur Motorway- Chauffeur Commuter- Chauffeur Universal- Chauffeur Urban Robot- Taxi Slide No. 19
Safety Impact Assessment of Automated Driving 6 Key results Motorway-Chauffeur can reduce 30 % of all accidents on German motorways at a market penetration of 50 %. This equals 2 % of all accidents on German roads. The Urban Robot-Taxi can avoid 26 % of all accidents with personal injury within city-limits at a market penetration of 50 %. This equals 17 % of all accidents on German roads. AD Ref However, there will be accidents remaining that automated vehicles cannot avoid (due to weather conditions or physics). But we can show that a human cannot avoid these accidents either. Slide No. 20
Piloting Automated Driving on European Roads L3Pilot Real World Data for Impact Assessment Large-scale Level 3 piloting 1,000 test drivers,100 vehicles in 11 European countries EC funded in Horizon 2020 34 partner Budget: 68 Mio., Funding: 36 Mio. Website: http://www.l3pilot.eu Slide No. 21
L3Pilot Evaluation Levels Europe Data Management Technical & Traffic Evaluation User evaluation Impact Evaluation Socio-economic impact evaluation Subjective data Objective driving data Frequency of driving scenarios DMs, PIs per driving scenarios e.g. Effect in Transition of control Long-term effects, usage Safety impact (e.g. avoided accidents) Environmental impact (e.g. fuel consumption) on European target level Slide No. 22
L3Pilot Evaluation Workflow Vehicles Obj. Data Subj. Data CDF Time series data Digital version Technical Evaluation Technical & Traffic Evaluation User & Acceptance Evaluation User Evaluation Performance Indicators per driving scenario Other inputs (e.g. accident & baseline data) Impact Assessment Estimated Impacts Socio-Economic Impact Assessment Impact Assessment Socio-Economic Impact 23 Slide No. 23
Summary Prospective safety impact assessment for automated driving requires new methodologies Automated driving provides many challenges with regards to impact assessment since limited real world data is available yet and many new aspects (e.g. user-interaction) needs to be taken into account Safety impact assessment shows positive results with different automation function Current research in L3Pilot start data collection for safety impact assessment Safety Impact Assessment in L3Pilot will provide results based on data from vehicles combined with simulation for the first time Slide No. 24
Contact THANK YOU FOR YOUR ATTENTION! Dr.-Ing. Adrian Zlocki QUESTIONS? fka Forschungsgesellschaft Kraftfahrwesen mbh Aachen Steinbachstr. 7 52074 Aachen Germany Phone Fax +49 241 80 25616 +49 241 8861 110 Email Internet zlocki@fka.de www.fka.de Slide No. 25