Conference Presentation, Lecture, Interview
Chris Gerdes (Stanford) on Technology, Policy and Vehicle Safety - MIT Self-Driving Cars
- Speaker Background: Chris Gerdes, a Stanford professor and former U.S. Department of Transportation (USDOT) Chief Innovation Officer (2016), now discusses autonomous vehicle (AV) technology and policy from a personal capacity.
- Research Vehicle "Shelly": An automated Audi TT capable of reaching 120 mph on race tracks, currently approaching the lap times of professional IndyCar drivers (J.R. Hildebrand).
- Performance Philosophy: The "Shelly" project utilizes physics-based calculations (force, mass, acceleration) to push tires to their friction limits, using machine learning to adapt to real-time variables like tire warm-up and track conditions.
- Safety Objective: The goal is to replicate human instinct for maximizing tire friction in emergency scenarios on public roads to enable the safest possible maneuvers when avoiding accidents.
- Regulatory Framework History: The current U.S. system relies on manufacturer self-certification for Federal Motor Vehicle Safety Standards (FMVSS), established by the 1966 National Traffic and Motor Vehicle Safety Act following Ralph Nader's "Unsafe at Any Speed."
- Certification Disparity: Unlike aviation, which requires pre-market government certification, the U.S. automotive sector allows vehicles to be sold upon manufacturer declaration of compliance.
- FMVSS Flexibility: NHTSA has issued a policy statement interpreting references to a "driver" in existing FMVSS to include "AI systems," allowing many automated vehicles to comply without new federal rules.
- Design Limitations: Vehicles lacking traditional controls (e.g., no steering wheel or pedals) currently require specific NHTSA exemptions as they do not meet existing physical equipment standards.
- Regulatory Lag: The formal rulemaking process for new safety standards takes 2–7 years, making it too slow to address rapid advancements in AI and autonomous driving technology.
- Federal Automated Vehicle Policy (Sept 2016): A voluntary 15-point safety assessment framework introduced to guide manufacturers without imposing rigid, premature standards.
- Operational Design Domain (ODD): Manufacturers must explicitly define the specific conditions (geography, weather, time of day) where their automated system is designed to operate.
- Fallback Conditions: Guidance requires manufacturers to define how a vehicle safely handles failure or exit from its ODD, ranging from human takeover to a "minimal risk condition" (e.g., stopping the car).
- Validation Methods: NHTSA accepts a mix of real-world testing, simulation, and analysis, noting that data from one region (e.g., Mountain View) may not transfer effectively to another (e.g., Cambridge).
- Ethical Frameworks: The policy encourages engineers to move beyond "trolley problems" toward practical risk reduction (e.g., barriers, warning signs) and established practices like Automatic Emergency Braking (AEB) that prioritize human safety over vehicle safety.
- Legal vs. Safety Conflict: Automated vehicles face dilemmas when legal codes (e.g., double yellow lines) conflict with safety needs (e.g., swerving to avoid a parked car or passing a cyclist).
- Human vs. Robot Driving: The speaker argues against programming AVs to mimic human error, which causes 94% of accidents, advocating instead for systems that exceed human performance capabilities.
- Data Sharing Strategy: Gerdes advocates for anonymized data sharing on edge cases (similar to the aviation ASIAS system) to accelerate neural network training without compromising intellectual property.
- Future Vehicle Design: Reduced accident rates from AVs could allow for "rollback" of passive safety features (airbags, heavy frames), potentially lowering vehicle mass and energy consumption to levels comparable to cycling.
- Liability: Liability for AV accidents is currently adjudicated through existing court systems using joint and several liability, with some manufacturers voluntarily accepting product liability.
- Global Policy Alignment: Discussions at the G7 Transportation Ministers' Meeting focused on harmonizing testing standards across different regulatory environments (e.g., U.S. self-certification vs. European pre-market testing).
- Open Source Viability: The voluntary guidance does not legally prevent open-source autonomous vehicles, provided a responsible entity signs the safety assessment and manages liability.