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

TL;DR

This paper introduces a resilient MPC approach for autonomous vehicles that uses adversarial disturbance sequences to enhance robustness in multi-vehicle traffic scenarios, outperforming traditional methods without prior training.

Contribution

The paper presents a novel adversarial disturbance-based MPC method that improves robustness in traffic scenarios by automatically generating adversarial predictions, reducing conservatism compared to existing approaches.

Findings

Successfully negotiates complex traffic situations in real-time

Outperforms nominal MPC and matches state-of-the-art reinforcement learning methods

Effectively prunes disturbance sequences to lower risk for the ego vehicle

Abstract

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