Resilient Branching MPC for Multi-Vehicle Traffic Scenarios Using Adversarial Disturbance Sequences

An approach to resilient planning and control of autonomous vehicles in multi-vehicle traffic scenarios is proposed. The proposed method based on model predictive (MPC), where alternative predictions the surrounding are determined automatically such that they intentionally adversarial ego vehicle. This provides robustness against inherent uncertainty predictions. To reduce conservatism, an assumption other agents no ill intent formalized. Simulation results from highway driving show real-time negotiates situations out scope for a nominal MPC performs favorably state-of-the-art reinforcement-learning approaches without requiring prior training. also effectively, with ability prune disturbance sequences lower risk