Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

2014 MATC Spring Lecture Series: Chris Schwarz

225 views

Published on

Towards Autonomous Vehicles

Published in: Engineering
  • Be the first to comment

  • Be the first to like this

2014 MATC Spring Lecture Series: Chris Schwarz

  1. 1. Towards Autonomous Vehicles Chris Schwarz National Advanced Driving Simulator
  2. 2. Acknowledgements • Mid-America Transportation Center – 1 year project to survey literature and report on state of the art in autonomous vehicles – Co-PI: Prof. Geb Thomas – Undergraduate students • Kory Nelson • Michael McCrary • Mathew Powell • Nicholas Schlarmann – http://matc.unl.edu/research/research_projects.php?researchID=405 – https://www.zotero.org/groups/autonomous_vehicles/items
  3. 3. Why Autonomous Vehicles? • Safety – 32,000 people killed each year, 93% due to driver error, billions in property damage – Autonomous vision is ‘crashless’ • Mobility – Safely increase traffic density (x2)-(x3) – Greater access for elderly, disabled, etc. • Sustainability – Fuel savings due to platooning (20%), eliminating traffic jams, reducing trip times, reducing ownership, reducing parking spaces
  4. 4. Cycles of Innovation
  5. 5. Vehicle Automation Partner Matrix Academic Government Private Military A G P M
  6. 6. An early experiment on automatic highways was conducted by RCA and the state of Nebraska on a 400 foot strip of public highway just outside Lincoln (“Electronic Highway of the Future - Science Digest (Apr, 1958)” 2013) A G P M
  7. 7. CMU NAVLAB • RALPH, ALVINN, YARF • In 1995, RALPH drove NAVLAB 5 over 3000 miles from Pittsburgh to Washington, DC. – Steered autonomously 96% of the way from Pittsburgh, PA to Washington DC Pomerleau, 1995, RALPH: Rapidly Adapting Lateral Position Handler, IEEE Symposium on Intelligent Vehicles, September, 1995 A G P M
  8. 8. National Automated Highway System A demonstration of the automated highway system in San Diego (1997). University of California PATH Program 1994-1997 A G P M
  9. 9. Intelligent Vehicle Initiative • Prevent driver distraction • Facilitate accelerated deployment of crash avoidance systems – Normal conditions • IVIS – Degraded condition • Visibility, drowsiness – Imminent crash • Rear end, lane depart, intersection, ESC 1997-2005 Multiple ADAS system. Image from IVBSS materials, courtesy of UMTRI Forward Crash Warning (FCW) Lateral Drift Warning (LDW) Lane-change/Merge (LCM) Curve speed Warning (CSW) Radar Vision A G P M
  10. 10. DARPA Grand Challenge Grand Challenge: 2004 – no winner 2005 – Stanley (Stanford) Urban Grand Challenge 2007 – Boss (CMU) A G P M
  11. 11. Connected Vehicles • DSRC (5.9 GHz) – Allocated in 2004 • Goals – Safety • Forward collision, intersection movement assist, lane change, blind spot, do not pass, control loss warning, emergency brake light warning – Mobility – Sustainability • AERIS 2004-present VII -> IntelliDrive -> Connected Vehicles Regulatory decision from NHTSA recently announced. V2V will eventually be required in new cars. A G P M
  12. 12. Google Self-Driving Car 2010 A G P M
  13. 13. NHTSA Automation Program • Licensing • Testing • Regulations • Cybersecurity • Currently recommends states only allow testing 2012-present Level Example Transition Time to Manual (Heuristic) 0 – No Automation Warning only -- 1 – Function-specific Automation ADAS < 1 second 2 – Combined Function Automation Super cruise < 1 minute 3 – Limited Self-Driving Automation Google car < 10 minutes 4 – Full Self-Driving Automation PRT -- NHTSA Levels of Automation A G P M
  14. 14. Future Societal Impacts Light Cars: A Virtuous Cycle Reduce mass Downsize engine Drivetrain brakes tires Smaller fuel supply Autonomous Car Sharing MIT’s Stackable City Car
  15. 15. A Bottom-up approach
  16. 16. Advanced Driver Assistance Systems ACC Pre-Crash LDWS Sensor Year Sensor Year Sensor Year Audi Radar/Video 2011 Camera 2007 BMW Camera 2007 Chrysler Laser 2006 Ford Radar 2009 Radar 2009 Camera 2010 GM Radar 2004 Camera 2008 Honda Radar 2003 Camera 2003 Kia Camera 2010 Jaguar Radar 1999 Lexus Laser 2001 Mercedes Radar 2001 Radar 2002 Camera 2009 Nissan Camera 2001 Saab Radar 2002 Toyota Laser 1998 Radar 2003 Camera 2002 Volkswagen Radar/Video 2011 Volvo Radar 2002 Radar/Video 2007 A 2011 review of commercial ADAS systems compares manufacturers, model year, and sensor type for three types of systems (Shaout, Colella, and Awad 2011)
  17. 17. ADAS Automation Abb. System Abb. System ESC Electronic Stability Control DD Drowsiness Detection FCW Forward Collision Warning AL Adaptive Lighting ACC Adaptive Cruise Control PM Pedal Misapplication LDW Lane Departure Warning TSR Traffic Sign Recognition LKA Lane Keeping Assist TJA Traffic Jam Assistant LCA Lane Change Assist CZA Construction Zone Assist RCTA Rear Cross Traffic Alert PA Parking Assistant BSD Blind Spot Detection PP Parking Pilot EBA Emergency Brake Assist HC Highway Chauffeur AEBS Advanced Emergency Braking System HP Highway Pilot ESA Emergency Steer Assist
  18. 18. A Top-down Approach
  19. 19. Personal Rapid Transit (PRT) • Fully autonomous • No operator, no controls • Low speed • May use a guideway • Morgantown PRT entered operation in 1975 in West Virginia
  20. 20. PRTs (cont.) • Morgantown, WV • Masdar City (on hold) • London Heathrow Airport • City Mobil 2 • Suncheon, South Korea • Punjab, India • Early criticisms of PRTs on guideways concern the scalability of the system • But new concepts are leaving guideways behind, alleviating some of these concerns
  21. 21. Elements of Automation
  22. 22. Automation Sensors High grade LIDAR Inconspicuous LIDAR GPS / IMU RADAR Cameras Digital MapsDSRC
  23. 23. Localization & Object Detection
  24. 24. Probabilistic Methods • The world is messy with uneven edges, bad lighting, poorly marked roads, and unpredictable people • Applications of probabilistic reasoning – Histogram filters (lane line tracking) – Particle filters, Kalman filters (object tracking) – Bayesian Networks (decision making) – Hidden Markov Models (state estimation)
  25. 25. Some Online Courses • Udacity online courses
  26. 26. Digital Maps & Mapping • Digital maps negate the need to dynamically map the environment • Simultaneous Localization & Mapping (SLAM) used to create environments in unmapped areas • Many modern path planning algorithms are based on A* algorithm • Must find the proper correspondence between the digital map and other sensor inputs
  27. 27. Challenges of Automation
  28. 28. Weather Challenges Bob Donaldson / Post-Gazette
  29. 29. Testing & Certification Logic Sensor Failures Kalman Filters False Positives Histogram Filters Particle Filters Data Fusion More data (images & video) More test cases Path Planning Decision Making Digital Maps All speeds Parking Lots Many more tests
  30. 30. Transfer of Control Transfer of Control to a Platoon Level Example Transition Time to Manual 0 – No Automation Warning only -- 1 – Function-specific Automation ADAS < 1 second 2 – Combined Function Automation Super cruise < 1 minute 3 – Limited Self-Driving Automation Google car < 10 minutes 4 – Full Self-Driving Automation PRT -- Example:
  31. 31. Legality • “Automated vehicles are probably legal in the United States” – Bryant Walker Smith • 1949 Geneva Convention on Road Traffic requires that the driver of a vehicle shall be at all times able to control it • Who is liable: the driver or the manufacturer? • California, Nevada, and Florida have paved the way with state laws for automated vehicles
  32. 32. Hacking Entry Points Entry point Weakness Telematics The benefit of such systems is that the car can be remotely disabled if stolen, or unlocked if the keys are inside. The weakness is that a hacker could potentially do the same. MP3 malware Just like software apps, MP3 files can also carry malware, especially if downloaded from unauthorized sites. These files can introduce the malware into a vehicles network if not walled off from safety-critical systems. Infotainment apps Car apps are like smartphone apps…they can carry viruses and malware. If the apps are not carefully screened, or if the car’s infotainment software is not securely walled off from other systems, then an attack can start with a simple app update. Bluetooth The system that connects your smartphone to your car can be used as another entry point into the in-vehicle network. OBD-II This port provides direct access to the CAN bus, and potentially every system of the car. If the CAN bus traffic is not encrypted, it is an obvious entry point to control a vehicle. Door Locks Locks are interlinked with other vehicle data, such as speed and acceleration. If the network allows two-way communication, then a hacker could control the vehicle through the power locks. Tire Pressure Monitoring System Wireless TPMS systems could be hacked from adjacent vehicles, identify and track a vehicle through its unique sensor ID, and corrupt the sensor readings. Key Fob It’s possible to extend the range of the key fob by an additional 30’ so that it could unlock a car door before the owner is close enough to prevent an unwanted entry.
  33. 33. Vehicle Networks to Secure Network Weakness LIN Vulnerable at a single point of attack. Can put LIN slaves to sleep or make network inoperable CAN Can jam the network with bogus high priority messages or disconnect controllers with bogus error messages FlexRay Can send bogus error messages and sleep commands to disconnect or deactivate controllers MOST Vulnerable to jamming attacks Bluetooth Wireless networks are generally much more vulnerable to attack than wired networks. Messages can be intercepted and modified, even introducing worms and viruses
  34. 34. Privacy • Electronic Data Recorders (Black Box) • Identified network traffic • De-identified data – The myth of anonymity • “Google’s self-driving car gathers almost 1 Gb per second” – Bill Gross, Idealab
  35. 35. Privacy By Design • Proactive not reactive • Privacy by default • Privacy embedded into the design • Full functionality (positive sum, not zero sum) • End-to-end security (full lifecycle protection) • Visibility and transparency • Respect for user privacy
  36. 36. Discussion
  37. 37. Case Study: Autonomous Intersections and Time to Collision Perception • Time to Collision (TTC) – range / range rate • Autonomous Intersection Management – U Texas at Austin – Reservation system Van der Horst, 1991 Autonomous Intersection (Top down) Autonomous Intersection (Driver's View)
  38. 38. The Trouble With Levels The evolution of vehicle automation and associated challenges • Levels are not a roadmap • Levels are not design guidelines • Levels discourage potentially helpful ideas like adaptive automation strategies
  39. 39. 5 – 30 years until autonomous vehicles hit the road

×