On Infusing Reachability-Based Safety Assurance within Probabilistic\n Planning Frameworks for Human-Robot Vehicle Interactions

Abstract

Action anticipation, intent prediction, and proactive behavior are all\ndesirable characteristics for autonomous driving policies in interactive\nscenarios. Paramount, however, is ensuring safety on the road --- a key\nchallenge in doing so is accounting for uncertainty in human driver actions\nwithout unduly impacting planner performance. This paper introduces a\nminimally-interventional safety controller operating within an autonomous\nvehicle control stack with the role of ensuring collision-free interaction with\nan externally controlled (e.g., human-driven) counterpart. We leverage\nreachability analysis to construct a real-time (100Hz) controller that serves\nthe dual role of (1) tracking an input trajectory from a higher-level planning\nalgorithm using model predictive control, and (2) assuring safety through\nmaintaining the availability of a collision-free escape maneuver as a\npersistent constraint regardless of whatever future actions the other car\ntakes. A full-scale steer-by-wire platform is used to conduct traffic weaving\nexperiments wherein the two cars, initially side-by-side, must swap lanes in a\nlimited amount of time and distance, emulating cars merging onto/off of a\nhighway. We demonstrate that, with our control stack, the autonomous vehicle is\nable to avoid collision even when the other car defies the planner's\nexpectations and takes dangerous actions, either carelessly or with the intent\nto collide, and otherwise deviates minimally from the planned trajectory to the\nextent required to maintain safety.\n