FOSP: Fine-tuning Offline Safe Policy through World Models
This work addresses safety constraints in vision-based robotic tasks for robotics and AI applications, presenting a novel approach to offline-to-online RL for safe generalization.
The paper tackles the problem of offline safe reinforcement learning struggling to generalize safely to unseen scenarios by proposing a method that fine-tunes an offline pretrained policy online using model-based RL and reachability guidance. The result is a significant improvement in generalization to unseen safety-constrained scenarios, validated on simulation benchmarks with five vision-only tasks and real-world robot deployment.
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.