Safe Motion Planning for Autonomous Driving using an Adversarial Road Model

This paper presents a game-theoretic path-following formulation where the\nopponent is an adversary road model. This formulation allows us to compute safe\nsets using tools from viability theory, that can be used as terminal\nconstraints in an optimization-based motion planner. Based on the adversary\nroad model, we first derive an analytical discriminating domain, which even\nallows guaranteeing safety in the case when steering rate constraints are\nconsidered. Second, we compute the discriminating kernel and show that the\noutput of the gridding based algorithm can be accurately approximated by a\nfully connected neural network, which can again be used as a terminal\nconstraint. Finally, we show that by using our proposed safe sets, an\noptimization-based motion planner can successfully drive on city and country\nroads with prediction horizons too short for other baselines to complete the\ntask.\n