Infrastructure for Automated Transport : State of the Practice in the U.S.

Size: px

Start display at page:

Download "Infrastructure for Automated Transport : State of the Practice in the U.S."

Marjory Robbins
5 years ago
Views:

2 Infrastructure for Automated Transport : State of the Practice in the U.S. Steve Dellenback, Ph.D., PMP Vice President R&D Southwest Research Institute San Antonio, Texas USA

3 State of the Practice 100 s of companies are working on some aspect of self driving technology: Technology companies OEMs Suppliers Basic premise: realizing AVs is much more difficult in mixed (manned and unmanned) environments Hype: The U.S. media is seriously over-hyping which is causing companies to over-claim (think: Tesla) Many claims are being made w/o supporting data Terminology is being abused (what is automated?) Cost of implementation being regularly ignored 3

4 SAE Autonomy Levels 4

Limitation of AV: Fallback to the Human (Moving from Level 3 to Level 4) Studies and incidents have shown that humans often come to trust a system very quickly.

5 Limitation of AV: Fallback to the Human (Moving from Level 3 to Level 4) Studies and incidents have shown that humans often come to trust a system very quickly. Once a person is actively disengaged from the driving process, it can take multiple seconds to regain situational awareness in order to take effective control. 5

6 Limitation of AV: Automation Anywhere, Anytime (Moving from Level 4 to Level 5) No system has encountered all possible scenarios. How do you quantify the capability to generalize to unknown scenarios? 6

State Level States With Self-Driving Car Legislation Many States have legislative, regulatory, and policy frameworks in place or in progress for: Connected and Automated Vehicles Autonomous Vehicles

7 State Level States With Self-Driving Car Legislation Many States have legislative, regulatory, and policy frameworks in place or in progress for: Connected and Automated Vehicles Autonomous Vehicles Autonomous Vehicle Legislation 18 states have passed legislation related to autonomous vehicles Governors in Arizona, Massachusetts, Washington and Wisconsin issued executive orders related to autonomous vehicles Source: 7

8 U.S. Policy Approaches Examples of State Approaches: California: Allows testing but requires registration with DMV and reporting of traffic incidents Florida: Requires no additional license, permit, or approval for driverless operations Michigan: Allows complete Autonomous operations Nevada: First jurisdiction to authorize use of AVs; requires special license for test vehicles Tennessee: Prohibited local governments from banning the use of AVs and established a per mile tax structure Louisiana: Defined terms (e.g. autonomous technology, driving mode, dynamic driving task) Alabama, North Dakota, and Utah: Authorized study committees Virginia: 2015 proclamation declared the Commonwealth "open for business"; 2017 established the Autonomous Systems Center of Excellence 8

9 Federal and State Responsibilities for CAV Readiness (from USDOT) 9

10 Other Perspectives Voiced by the U.S. Government Comment commonly made that automated cannot happen without connected : USDOT/FHWA focuses on CAV (Connected Automated Vehicle) CV needs to converge with AV by 2025: They need to start the research on AV now Data: the infrastructure has significant data to assist the AV effort: AVs have data that would be valuable to the infrastructure management Would like to see federal AV policy by

11 What Timeline is USDOT/FHWA Discussing Editorial comments: CV (heavily USDOT funded) has been 15+ years in the making and still incomplete AV (almost exclusively privately funded) is working at market speed 11

12 As of December 28, 2017, California DMV has issued Autonomous Vehicle Testing Permits to the following (Listed by Date issued) 1. Volkswagen Group of America 2. Mercedes Benz 3. Waymo 4. Delphi Automotive 5. Tesla Motors 6. Bosch 7. Nissan 8. GM Cruise LLC 9. BMW 10. Honda 11. Ford 12. Zoox, Inc. 13. Drive.ai, Inc. 14. Faraday & Future Inc. 15. Baidu USA LLC 16. Valeo North America, Inc. 17. NIO USA, Inc. 18. Telenav, Inc. 35. Almotive Inc 19. NVIDIA Corporation 36. Aurora Innovation 20. AutoX Technologies Inc 37. Nullmax 21. Subaru 38. Samsung Electronics 22. Udacity, Inc 39. Continental Automotive Systems Inc 23. Navya Inc. 40. Voyage 24. Renovo.auto 41. CYNGN, Inc 25. UATC LLC (Uber) 42. Roadstar.Ai 26. PlusAi Inc 43. Changan Automobile 27. Nuro, Inc 44. Lyft, Inc. 28. CarOne LLC 45. Phantom AI 29. Apple Inc. 46. Qualcomm Technologies, Inc. 30. Bauer s Intelligent Transportation47. aipod, Inc. 31. Pony.AI 32. TuSimple 33. Jingchi Corp 48. SF Motors Inc. 49. Toyota Research Institute 34. SAIC Innovation Center, LLC 12

13 California DMV Autonomous Vehicle Disengagement Reports Summary Developer # of Vehicles Miles Driven Disengagements Miles between Disengagements Disengagement Per 1,000 miles 2015 Bosch Delphi 2 16, GM Cruise? Google , , Nissan 4 1, Mercedes Benz 2 1,739 1, Telsa no testing no testing no testing no testing no testing 2015 Volkswagon 2 5, BMW? Bosch , Delpi 2 3, Ford GM Cruise 25 9, Google / Waymo , , Honda no testing no testing no testing 2016 Nissan NA 5 4, Mercedes Benz? Tesla Volkswagon no testing no testing no testing

Challenge: Perception Issues The farther away a vehicle can perceive the environment, the more capability it will have to maneuver safely at higher speeds.

14 Challenge: Perception Issues The farther away a vehicle can perceive the environment, the more capability it will have to maneuver safely at higher speeds. Some sensing modalities are better at ranging, some better at context, and some are better in degraded weather conditions. In addition, utilizing multiple perception technologies naturally provides robustness to specific failure modes of a single sensor. 14

Challenge: Infrastructure Markings Lane line detection has become more prevalent in available systems for lane departure warning, lane keep assist and vehicle automation.

15 Challenge: Infrastructure Markings Lane line detection has become more prevalent in available systems for lane departure warning, lane keep assist and vehicle automation. Often, these systems rely on relatively pristine lane markings and can often fail in the presence of faded or ambiguous lines or other markers (such as Botts dots or other raised reflective markers). 15

Traffic signals have proven to be more difficult: their configurations vary widely,

16 Challenge: Infrastructure Signage Detection of road signage has greatly improved in recent years, interpreting speed limit signs, stop signs, and others. Traffic signals have proven to be more difficult: their configurations vary widely, they often appear in cluttered visual environments, and they are often smaller visually than other signage. 16

Challenge: Adverse Weather Conditions Visible-spectrum cameras same problems that a human driver would: their range and effectiveness are severely limited in heavy rain, snow, or dense fog.

17 Challenge: Adverse Weather Conditions Visible-spectrum cameras same problems that a human driver would: their range and effectiveness are severely limited in heavy rain, snow, or dense fog. The laser pulses used by LIDAR sensors can reflect off raindrops, snowflakes, or other particles, surrounding the vehicle in false positives. Radar is least affected by most of the adverse conditions that afflict the visible-spectrum sensors, but has the significant drawback of reduced resolution. 17

Newer techniques, such as deep learning,

18 Challenge: Sensor Processing and Generalization How do you train and verify that a system will work in every scenario that it will encounter? Newer techniques, such as deep learning, seem to broaden the ability to generalize processing in various scenarios, but how to you validate what is inside if it can be a black box? 18

19 Challenge: Understanding Intent of Other Drivers and Pedestrians What is this guy trying to communicate? Eye contact Body (vehicle) language Derive intent 19

20 Challenge: Weather What is the common visual theme when you see images of driverless cars that are commonly shared? Bad weather: Occludes highways / changes appearance of objects Vehicle dynamics 20

Efforts at Automated Driving have focused on: Paved roads Pavement markings / traffic control devices To finish the drive in

21 Challenge: How Do You Finish The Drive Many higher level automation solutions currently rely on a priori, high definition maps for localization. Efforts at Automated Driving have focused on: Paved roads Pavement markings / traffic control devices To finish the drive in many places you need: Ability to navigate roads that are not paved or mapped (63% of world wide roads paved, 65% in the US): US: 2,744,171 miles of paved road, 1,421,083 miles of unpaved Environments with lots of vegetation and no stripes/signs

computers TPMS are 4 computers and wireless >100 million lines

22 Next Threat to AV in the US: Cybersecurity Cars are becoming complex 1965: No computers No software 2017: Up to ~200 computers TPMS are 4 computers and wireless >100 million lines of code Connected to the Internet 22

23 Cybersecurity: Connected Environment For CAV to be successful it must be integrated into the transportation infrastructure 23

Cybersecurity: Vehicle CAN bus replaced point to point wiring, all interconnected: Messages are broadcast Messages are easily cracked ECU broadcast any message engine electronics, transmission

24 Cybersecurity: Vehicle CAN bus replaced point to point wiring, all interconnected: Messages are broadcast Messages are easily cracked ECU broadcast any message engine electronics, transmission electronics, chassis electronics, active safety, driver assistance, passenger comfort, entertainment systems, electronic integrated cockpit systems, lambda control, on-board diagnostics, cooling system control, ignition system control, turbo charger control, lubrication system control, fuel injection rate control, throttle control, transmission electronics, chassis electronics, anti-lock braking system, traction control system, electronic brake distribution, electronic stability program, passive safety, air bags, hill descent control, emergency brake assist system, driver assistance, lane assist system, speed assist system, blind spot detection, park assist system, adaptive cruise control system, pre-collision assist, passenger comfort, automatic climate control, electronic seat adjustment with memory, automatic wipers, automatic headlamps - adjusts beam automatically, automatic cooling - temperature adjustment, infotainment systems, navigation system, vehicle audio, information access, tire pressure monitoring system (Wikipedia) 24

Impact of these attacks: Erodes public trust Raises awareness improves security practices

25 Cybersecurity: Possible Penetration Effects ECU control via CAN bus messages Auto start/stop Brakes Lane assist / collision avoidance Cruise control Steering Software Over the Air updates Impact of these attacks: Erodes public trust Raises awareness improves security practices Financial cost (recalls, lawsuits, etc) Image Source: Wired Magazine Image Source: Wired Magazine 25

26 Punchline: Perception/Behaviors are Challenging 95% is easy, it s the last 5%... Deer in the headlights Realistic driving June 2014 in DC Taxi strike How to nose into traffic 26

27 Automating a Conscious Computers needing to make value decisions For years researchers have been trying to teach computers to learn: Computing power can now support Convolutional (or deep) neural networks provide a framework Consider the unforeseen challenges you have encountered how to prepare a computer for all of them Classic example: Trolley problem: Graphic Source: 27

28 Looking out to the Horizon: What is Next? (presenter s personal opinion) Next 3 to 20 years: Don t expect to see automated vehicles regularly used on mixed use public roads Military operations that can accept collateral damage Closed operations (such as mining, agriculture, ports, retirement communitities) have less unpredictability: No teenage drivers Limited obstacles Very well known environment (that does not change much) Potential game changer: dedicated transit or truck or technology lanes Holy grails: Perception (sensors) Cost Cybersecurity Use of technology : generational 28

29 Questions / Comments? Steve Dellenback sdellenback@swri.org

California s Autonomous Vehicles Testing Progam

California s Autonomous Vehicles Testing Progam July 24, 2018 Bernard Soriano, Ph.D. Deputy Director California Department of Motor Vehicles Why Autonomous Vehicles? Safety o 37,461 people killed in crashes