Justifying Emergency Drift Control for Automated Vehicles

Abstract

Expert human drivers can execute emergency steering actions to avoid sudden events like a deer crossing the road. However, justifying beyond-the-limit emergency maneuvering for automated driving systems is exceptionally challenging. Emergency maneuvering often requires non-linear control policies without stability guarantees. Liability concerns, ethics, lack of safety guarantees, and non-linear system dynamics convolute an already complicated problem. Against this backdrop, we propose a principled approach to justify a particular type of emergency steering in safety-critical situations. A limit-handling controller is justified and deployed to execute the emergency maneuver upon a conventional controller's formally verified incapability to handle. We claim this check justifies the execution of the emergency maneuver as we show failure is mathematically inevitable otherwise. The simulation-based experimental validation shows that using backward reachability analysis, the proposed approach can determine emergencies. The validation justifies using limit-handling controllers for collision avoidance in a scenario where the baseline controllers fail catastrophically.