Towards Responsibility-Sensitive Safety of Automated Vehicles with Reachable Set Analysis

Abstract

One of the most critical, unsolved challenges for the introduction of automated vehicles is safety verification of planned trajectories. The most promising concepts approaching this topic are worst-case occupancy predictions based on reachable set analysis and the definition of Responsibility-Sensitive Safety (RSS) that formalizes dangerous situations and proper responses. Worst-case predictions result in over-conservative behavior while the RSS approach makes strong assumptions w.r.t. lateral and longitudinal behavior. Our contribution is a first step in bringing both worlds together, to benefit from respective advantages. First, we define RSS-motivated safe-states for merge and crossing scenarios, that ensure absolute safety towards leading vehicles, appropriate time gaps towards following vehicles in merging lanes and minimum clearance time of conflict zones with crossing lanes. We then show how to integrate these safety constraints in a trajectory and behavior planner using reachable sets and finally illustrate its usefulness in various simulative evaluations.