Offline Safe Reinforcement Learning Using Trajectory Classification

TL;DR

This paper introduces a trajectory classification approach for offline safe reinforcement learning, enabling policies to generate desirable behaviors while avoiding unsafe ones, thus improving safety and reward outcomes.

Contribution

The paper proposes a novel offline safe RL method that classifies trajectories into desirable and undesirable sets, bypassing complex min-max optimization and enhancing safety and performance.

Findings

Outperforms baseline methods on DSRL benchmark

Achieves higher rewards and better safety constraints

Effectively distinguishes safe and unsafe trajectories

Abstract

Offline safe reinforcement learning (RL) has emerged as a promising approach for learning safe behaviors without engaging in risky online interactions with the environment. Most existing methods in offline safe RL rely on cost constraints at each time step (derived from global cost constraints) and this can result in either overly conservative policies or violation of safety constraints. In this paper, we propose to learn a policy that generates desirable trajectories and avoids undesirable trajectories. To be specific, we first partition the pre-collected dataset of state-action trajectories into desirable and undesirable subsets. Intuitively, the desirable set contains high reward and safe trajectories, and undesirable set contains unsafe trajectories and low-reward safe trajectories. Second, we learn a policy that generates desirable trajectories and avoids undesirable trajectories, where (un)desirability scores are provided by a classifier learnt from the dataset of desirable and undesirable trajectories. This approach bypasses the computational complexity and stability issues of a min-max objective that is employed in existing methods. Theoretically, we also show our approach's strong connections to existing learning paradigms involving human feedback. Finally, we extensively evaluate our method using the DSRL benchmark for offline safe RL. Empirically, our method outperforms competitive baselines, achieving higher rewards and better constraint satisfaction across a wide variety of benchmark tasks.