Conference Presentation, Lecture, Interview
Chris Gerdes (Stanford) on Technology, Policy and Vehicle Safety - MIT Self-Driving Cars
- Federal rulemaking for new safety standards may become ineffective within approximately two to seven years due to regulatory pacing relative to current automated vehicle technology.
- The Federal Automated Vehicle Policy, introduced in September, is expected to transition from voluntary guidance into binding standards and laws as the industry adopts the framework.
- Future society may roll back passive safety requirements like airbags and seatbelts if vehicles achieve extremely low accident rates and full automation.
- Significant vehicle mass reduction could occur in the future, potentially making transportation comparable to cycling environmentally, as safety becomes decoupled from high-speed crash-worthiness.
- Testing focus will shift from extreme performance capabilities to mundane factors like driving patterns once machine and human performance differentials narrow.
- Public and government discussions will occur to define an acceptable safety level if industry standards fail to meet public expectations.
- Visual mechanical failures in automated vehicles may differ from human error, potentially raising public fear despite a decline in overall national accident statistics.
- Accelerated data sharing regarding edge cases and extreme situations will enhance neural network training and vehicle development.
- The aviation industry's ASIAS system, currently utilized by 40 airlines for anonymous safety data, may serve as a model for the automated vehicle ecosystem to build trust and facilitate data sharing.
- Expanding operational design domains from hyperlocal to nationwide coverage will increase the necessity for common data sets and standardized testing protocols, though no dataset can capture all eventualities.
- Regulators will continue refining their understanding of operational design domains, requiring manufacturers to explain how their systems handle specific defined limits.
- Open source development remains feasible under current guidance provided a responsible party assumes liability for the safety assessment of specific software versions.
- The National Highway Traffic Safety Administration is anticipated to continue pilot or prototype work on standardized data sharing for maps, accidents, and simulations.
- State-level liability rules may remain divergent despite initial industry commitments from companies like Volvo and Google regarding product liability for accidents.
- Further information regarding government work on standardized testing and data sharing protocols is expected in the not-so-distant future.
- Early engagement between the government and manufacturers will replace the approach of waiting for premature regulations to be enacted.
- Conservative or separated driving environments may be developed to further reduce the need for heavy crash-worthiness standards.
- Best practices are expected to evolve from the collective efforts of companies and universities rather than being solely developed by the National Highway Traffic Safety Administration.