COptiDICE: Offline Constrained Reinforcement Learning via Stationary Distribution Correction Estimation
This addresses safety-critical applications where direct environment interaction is risky, though it appears incremental as it builds on existing offline RL methods with a focus on constraint handling.
The paper tackles the offline constrained reinforcement learning problem by developing COptiDICE, an algorithm that optimizes policies in the stationary distribution space to maximize return while satisfying cost constraints from a pre-collected dataset, and it outperforms baselines in constraint satisfaction and return-maximization.
We consider the offline constrained reinforcement learning (RL) problem, in which the agent aims to compute a policy that maximizes expected return while satisfying given cost constraints, learning only from a pre-collected dataset. This problem setting is appealing in many real-world scenarios, where direct interaction with the environment is costly or risky, and where the resulting policy should comply with safety constraints. However, it is challenging to compute a policy that guarantees satisfying the cost constraints in the offline RL setting, since the off-policy evaluation inherently has an estimation error. In this paper, we present an offline constrained RL algorithm that optimizes the policy in the space of the stationary distribution. Our algorithm, COptiDICE, directly estimates the stationary distribution corrections of the optimal policy with respect to returns, while constraining the cost upper bound, with the goal of yielding a cost-conservative policy for actual constraint satisfaction. Experimental results show that COptiDICE attains better policies in terms of constraint satisfaction and return-maximization, outperforming baseline algorithms.