Navigation Safety for Autonomous Vehicles
Autonomous vehicles can make air, ground, and sea transportation safer, more cost-efficient and energy-efficient. These benefits are predicated on autonomous navigation systems achieving unprecedented levels of safety. Substantial resources have been deployed over the past 30 years in civilian aviation to meet limits of acceptability on positioning errors, or ‘alerts limits’, of 10 m to 35 m with high probability (on the order of 1-10-9). In comparison, alert limits for self-driving cars will range from 0.5m to 1m, which will only be achievable using new multi-sensor systems.
This presentation describes the design, analysis and evaluation of new methods to quantify navigation safety for autonomous vehicles. The first part of the talk provides an overview of Advanced Receiver Autonomous Integrity Monitoring (ARAIM), a cooperative research effort by the European Union and the United States to analyze future Global Navigation Satellite Systems (GNSS). Research contributions include, for example, the design of optimal estimators and detectors, which specifically minimize safety risk rather than maximizing accuracy. The second part of the presentation investigates whether analytical safety methods used in aviation can be leveraged for autonomous ground vehicle navigation. Self-driving cars recently caused several accidents, including the May 2016 crash of a Tesla Model S that caused a fatality on a Florida highway. The presentation will introduce a new concept for quantifying risks involved in data association, a process aimed at matching sensor measurements with mapped landmarks, which is needed for laser and radar localization.