RoaD: Rollouts as Demonstrations for Closed-Loop Supervised Fine-Tuning of Autonomous Driving Policies

TL;DR

RoaD is a novel method that uses a policy's own closed-loop rollouts, guided by experts, as additional training data to improve autonomous driving policies, reducing data needs and enhancing robustness.

Contribution

We introduce RoaD, a simple, efficient closed-loop fine-tuning method that leverages rollouts as demonstrations with expert guidance, outperforming prior methods in autonomous driving.

Findings

Improves driving score by 41% in AlpaSim

Reduces collisions by 54% in AlpaSim

Performs comparably or better than prior CL-SFT methods on WOSAC

Abstract

Autonomous driving policies are typically trained via open-loop behavior cloning of human demonstrations. However, such policies suffer from covariate shift when deployed in closed loop, leading to compounding errors. We introduce Rollouts as Demonstrations (RoaD), a simple and efficient method to mitigate covariate shift by leveraging the policy's own closed-loop rollouts as additional training data. During rollout generation, RoaD incorporates expert guidance to bias trajectories toward high-quality behavior, producing informative yet realistic demonstrations for fine-tuning. This approach enables robust closed-loop adaptation with orders of magnitude less data than reinforcement learning, and avoids restrictive assumptions of prior closed-loop supervised fine-tuning (CL-SFT) methods, allowing broader applications domains including end-to-end driving. We demonstrate the effectiveness of RoaD on WOSAC, a large-scale traffic simulation benchmark, where it performs similar or better than the prior CL-SFT method; and in AlpaSim, a high-fidelity neural reconstruction-based simulator for end-to-end driving, where it improves driving score by 41\% and reduces collisions by 54\%.