Lexicographic Mixed-Integer Motion Planning with STL Constraints

Autonomous vehicles are subject to various constraints, such as following the rules of the road (ROTR), adhering to schedules, or providing a comfortable driving experience. However, realizing a driving behavior complying with all constraints is challenging since it is not always possible to satisfy them simultaneously, necessitating the formulation of compromises. In this paper, we propose a solution to this challenge by decomposing the specification of an autonomous vehicle into a rulebook utilized by a novel optimization-based minimum - violation motion planner. In particular, our planner uses reachable sets to prevent collisions with other road users, and it minimally violates the ROTR formalized in signal temporal logic (STL). Furthermore, a mixed-integer convex program (MICP) realization of the planner is provided to demonstrate its effectiveness, especially in dynamically changing environments. We evaluate our approach using realistic ROTR on 1780 scenarios from the CommonRoad benchmark suite. Our results show that our planner generates safe and feasible trajectories, indicating its potential for real-world applications.