FOSP: Fine-tuning Offline Safe Policy through World Models

TL;DR

This paper introduces FOSP, a method combining offline safe RL with online fine-tuning using world models to enhance safety and generalization in vision-based robotic tasks, validated through simulations and real-world experiments.

Contribution

It presents a novel offline-to-online safe RL framework that employs model-based RL and reachability guidance for improved safety and generalization in robotic tasks.

Findings

Significant improvement in safety generalization on unseen scenarios.

Effective online fine-tuning after offline pretraining.

Validated on five vision-only simulation tasks and real robots.

Abstract

Offline Safe Reinforcement Learning (RL) seeks to address safety constraints by learning from static datasets and restricting exploration. However, these approaches heavily rely on the dataset and struggle to generalize to unseen scenarios safely. In this paper, we aim to improve safety during the deployment of vision-based robotic tasks through online fine-tuning an offline pretrained policy. To facilitate effective fine-tuning, we introduce model-based RL, which is known for its data efficiency. Specifically, our method employs in-sample optimization to improve offline training efficiency while incorporating reachability guidance to ensure safety. After obtaining an offline safe policy, a safe policy expansion approach is leveraged for online fine-tuning. The performance of our method is validated on simulation benchmarks with five vision-only tasks and through real-world robot deployment using limited data. It demonstrates that our approach significantly improves the generalization of offline policies to unseen safety-constrained scenarios. To the best of our knowledge, this is the first work to explore offline-to-online RL for safe generalization tasks.