Dutch Automated Vehicle Initiative The need for on-road assessment of the human interaction with automated driving Verkehr Aktuell, 23 April 2015, München R. Happee, TU Delft, Visiting Professor TU München r.happee@tudelft.nl / www.davi.connekt.nl Dutch Automated Vehicle Initiative 2015 1
Contents DAVI Hands free driving Full automation The Dutch invitation for on road testing Discussion Dutch Automated Vehicle Initiative 2015 2
Hands free Cruise Control with automated steering Highway in 2020 Capable driver Full Automation No steer & pedals User only has to select the destination Mixed with normal traffic Dutch Automated Vehicle Initiative 2015 3
Is it safe? Hands Free demonstrated on public roads with experts/developers Full automation operational in closed environments first pilots in mixed traffic Environment perception <100% detection of all road users in all weather and traffic conditions Will we accept any accidents with Automated Vehicles? Dutch Automated Vehicle Initiative 2015 4
Dutch Automated Vehicle Initiative To investigate, improve and demonstrate automated driving on public roads Assess & improve technology Study human behaviour Prove safety Pursue legalisation Create public awareness Dutch Automated Vehicle Initiative 2015 5
Levels of Automation Partial Hands & feet off Eyes on the road Prepared to resume control any time High (SAE 3-4) Hands & feet off Eyes off Engage in other activities Prepared to resume control when requested Full Automation (SAE 5) Program your destination No drivers, only users Dutch Automated Vehicle Initiative 2015 7
Safety by redundancy Vehicle technology Perception Control Software reliability Infrastructure Up to date maps Road markings Collision warning Traffic management Human Takeover Driver / controller can always overrule Misuse Safe stop Disrupt traffic Dutch Automated Vehicle Initiative 2015 8
Safety by restriction Monitor driver state Enforce hands on the steering wheel Enforce eyes on the road Awake and on driver seat Restrict conditions Limit operation speed Clear sight only Highway only Well marked roads (certified?) Support & Partial Automation High Automation Consumer shall be explicitly informed of restrictions Automation can request or force the driver to resume control when used beyond these restrictions Dutch Automated Vehicle Initiative 2015 9
DAVI projects & partners DAVI open initiative Human Factors of Automated Driving (HFAuto) Marie Curie FP7 Truck Merging Support STW-HTSM From Individual Automated Vehicles to Cooperative Traffic Management (IAVTRM) STW-OTP Standardized Self-diagnostic Sensing Systems for Highly Automated Driving (S4-DRIVE) STW-OTP WEpods Full Automation TU Delft, RDW, TNO, Connekt, SWOV, UT, Fontys, Hanze, HAN, RUG, Toyota Motor Europe, NXP, Imtech, VisLab (IT), SKF, Technolution, Almende, V-tron, Trinité Mapscape, DLR, TRL, ITS-Leeds TU Delft, TU München, Univ of Southampton, Univ of Twente, Chalmers Univ of Technology, IFSTTAR (France), VTI (Sweden), Volvo Truck, Volvo Car, BMW, Jaguar, Toyota Motor Europe, Continental, TNO, SWOV TU Delft, TU/e, DAF, SKF, TNO, NXP TU Delft, Toyota Motor Europe, TNO, NXP, Imtech Traffic & Infra, RDW, Connekt, SWOV, Technolution, Almende, V-tron, Trinité Automation, VisLab TU Delft, NXP, TNO, Toyota Motor Europe, Tata Motors Europe, Technolution, InnoSenT, Melexis, VisLab Province of Gelderland Dutch Automated Vehicle Initiative 2015 10
Demo Dutch N11 15-19 June 1998 Dutch Automated Vehicle Initiative 2015 11
DAVI DEMO November 2013 Innovatie Estafette RAI & A10 Amsterdam Automated Car Following (CACC) Radar + Camera + Communication Automated Steering Camera + Radar Automated gap creation Communication + Radar + Camera No dedicated infrastructure Dutch Automated Vehicle Initiative 2015 12
FP7-PEOPLE-2013-ITN #605817 3.8M, November 2013 October 2017 Coordinator - TU Delft, Netherlands Netherlands, Germany, Sweden, France, UK http://hf-auto.eu r.happee@tudelft.nl Happee R. Human Factors of Automated Driving, Presented at TRB, 14 Jan 2015, in the Dutch Automated Vehicle Initiative 2015 13 Human Factors of Road Vehicle Automation Joint Subcomittee of AND10, AHB15, AHB30
HFAuto objectives 1. New knowledge about human capabilities and side effects 2. Multimodal human-machine interface linked to a driver-state monitor 3. Traffic simulation models predicting effects of highly automated driving on traffic flow, safety, and fuel efficiency 4. Linking to roadmaps for market introduction of automated driving, including legislation 5. Trained researchers to address human factors challenges of automated driving Dutch Automated Vehicle Initiative 2015 14
13 PhD Candidates & 1 Post Doc Literature reviews Surveys (Opinions) Driving Simulator Studies Traffic Flow Simulations On Road Experiments FOT & Naturalistic (ACC) Dutch Automated Vehicle Initiative 2015 15
Highly Automated Driving SAE level 3&4 Safety Optimise Situation & Mode Awareness Prove that humans can regain control, even after reading & texting Quantify reaction time and accuracy after transitions Benefit Assessment Quantify human behaviour (usage, speed, distance) Traffic Flow Simulation Attractiveness Redefine the joy of driving Dutch Automated Vehicle Initiative 2015 16
Human Factors Knowledge ACC & HAD 1 Many studies on ACC Driving simulators Field Operational Tests Naturalistic Driving Quite some studies on (H)AD Extensive simulator studies Studies on public roads always with expert drivers Davis et al. 2008; Schieben & Flemish 2010) and even with a police escort (Lank et al. 2011). 1. de Winter JCF, Happee R, Martens MA, Stanton NA. (2014). Effects of ACC and highly automated driving on workload and situation awareness: A review of the empirical evidence. TRPF: Traffic Psychology and Behaviour. Dutch Automated Vehicle Initiative 2015 17
Driving simulator validity How valid are acceptance, trust and behaviour? Artificial experimental conditions Mostly < 60 minutes including training No or limited motion Unrealistic surrounding traffic Not using state of the art traffic flow models We don t know the reaction of other road users to automated vehicles Road to simulator comparisons show relative validity for manual driving For automated driving we don t know Dutch Automated Vehicle Initiative 2015 18
Workload Questionnaires (TLX 30 studies*) Secondary task (visual 12 studies*) Workload (%) Secondary Task Performance (%) 50 300 40 250 30 20 200 150 100 10 50 0 Control ACC HAD 0 Control ACC HAD * de Winter 2014 Dutch Automated Vehicle Initiative 2015 20
Risk (%) Perceived risk (%) 80 70 60 50 40 30 20 10 Acceptance and Trust Too much trust drivers fail to override the system when it is faulty automation complacency Lack of trust automation will not be sold or used short following distances will not be selected Clear automated Clear manual Fog automated Fog manual Saffarian Ergonomics (2012) 0 0 20 40 60 80 100 120 140 160 180 Distance (m) Dutch Automated Vehicle Initiative 2015 21
Critical Events HAD and ACC evoke long response times and an elevated rate of (near-) collisions in critical events If the automation fails unexpectedly with very little time for the human to respond almost all drivers crash If drivers receive a timely warning almost all drivers safely avoid collisions * de Winter 2014 Dutch Automated Vehicle Initiative 2015 22
Takeover Request eyes off the road End of lane marking End of lane Take Exit Eyes off the Road No errors 4 seconds sufficient Takeover indicated by tone 4-8 s before need to act Errors reduce with warning time (no/wrong action) Errors using mirrors remain Dambock PhD 2013 TU Munich Dutch Automated Vehicle Initiative 2015 23
Takeover Request eyes off road Warning sound / light / seat vibration 0.5s Eye movement onset 1.1s Eyes on road Rebuild environment awareness 2.5s Hands on wheel / feet on pedals Take control 30s Regain precision Dambock PhD TU Munich 2013 Dutch Automated Vehicle Initiative 2015 24
Human machine interface Published ACC and HAD studies use simple interfaces abrupt transitions simple control algorithms Well designed interfaces may enhance mode awareness & trust takeover performance with gradual transfer of control Zimmerman 2014 Flemisch Dutch Automated Vehicle Initiative 2015 26
Traffic Flow Simulation Current models predict >double highway capacity Even 10% penetration is relevant We need to quantify When automation will be used Selected speeds and distances Behaviour at transitions Will drivers engage in other tasks? Will they drive more often, and drive in congested hours? Preferably with non-expert drivers on public roads Dutch Automated Vehicle Initiative 2015 27
Automated lane changing & merging Shown in movies carmakers Proof of safety not found DAVI simulation studies Optimize gap & timing 1 Optimize collective benefits 1 Trajectory control with non-linear vehicle model 2 Next Sensors fusion & inaccuracy Role of the driver? 1. Wang M, Hoogendoorn SP, Daamen W, van Arem B, Happee R. Novel predictive approach for unified lanechanging and car-following control. Submitted 2. Gottardis P, Manazza SS. Automated Controlled Vehicle Based on Non-Linear Model Predictive Control Connected to a Safety Path Planner with Online Collision Risk Estimation MSc thesis TUDelft / Polimi 3 Oct 2014 Dutch Automated Vehicle Initiative 2015 28
Hands free driving outlook Multimodal human-machine interface linked to driver-state monitor Quantify ability to resume control Prolonged driving Engaging in other task In simulators and on public roads Assess trust and use on public roads Traffic simulation models predicting effects on traffic flow, safety, and fuel efficiency Dutch Automated Vehicle Initiative 2015 29
WEpods Full Automation Dutch Automated Vehicle Initiative 2015 30
Full Automation (SAE level 5) Private / shared / public Mobility elderly, kids(?) On demand Door to door Door to station Self-parking or drive to other user 24 hr per day (supervision needed) Optimal size (efficient) Major impact on multimodal mobility & infrastructure Stations / hubs Parking Dutch Automated Vehicle Initiative 2015 31
WEpods Gelderland NL The first selfdriving vehicles in the Netherlands will drive in Ede and Wageningen (Gelderlander 1 Oct 14). Innovation From closed track to public roads Safe interaction with pedestrians, cyclists, cars On demand public transport Safety by Low speed Conservative control strategies Control room monitoring Dutch Automated Vehicle Initiative 2015 32
WEpods EZ10 vehicle Citymobil2 4 lasers add camera & radar & detailed map fixed route 11km studied in depth Control Room Dutch Automated Vehicle Initiative 2015 33
Outlook full automation Towards more complex traffic Reliably detect pedestrians & cyclists Predict the envelope of possible future behavior (intent recognition) Interact (replace eye contact) Redefine the transport system On demand public transport Door to door Door to train Dutch Automated Vehicle Initiative 2015 34
Dutch invitation for on-road testing Dutch Automated Vehicle Initiative 2015 35
Dutch Automated Vehicle Initiative 2015 36
Dutch Automated Vehicle Initiative 2015 37
Exemption for large-scale testing Four steps: 1. Intake: Mission analysis Functional description Comprehensive risk analysis EMC 2. Desk Research: In consultation with the applicant. 3. Testing on a closed proving ground: Physical inspection of the vehicle(s) Happy flow testing Stress testing 4. Limited admittance with an exemption: With requirement and conditions, for example insurance, test times, test location, duration, monitoring and logging, gfeddes@rdw.nl Dutch Automated Vehicle Initiative 2015 38
Road, infrastructure and behavior Applicant RDW RDW Road Authority Behavoir Intake Deskresearch Vehicle Research Road(s) Research Road(s) Proving Ground Observation Observation Exemption Additional Requirements Additional Risks Use of exemption Evaluation with all parties involved Dutch Automated Vehicle Initiative 2015 39
Automation & ADAS Sharing the same sensors, tracking, fusion ADAS already rewarded by consumer tests AEB provides temporary automation How to deal with uncertainty? Automation shall react reliably in all critical events an early reaction will reassure other road users ADAS may miss some critical events false detections or early reactions annoy the driver Dutch Automated Vehicle Initiative 2015 40
OUTLOOK Automated Driving Hands Free Driving Licenced driver behind the wheel Mixed traffic Individual & platooning Highway 2020 Safe stop needed? Full Automation Initially low speed From closed tracks to public roads/space Prove safety Legislation Dutch Automated Vehicle Initiative 2015 41
Benefits Hands Free & Full Automation Minimize accidents (2% of GDP) Double highway capacity Reduce fuel consumption (20%) Free up time now lost in traffic Full Automation Enhanced mobility elderly, kids(?) Major impact on multimodal mobility Dutch Automated Vehicle Initiative 2015 42