Explaining Reinforcement Learning Policies through Counterfactual Trajectories
This work addresses the need for interpretability in reinforcement learning to help human developers assess agent reliability in real-world scenarios, though it is incremental as it builds on existing trajectory-based methods.
The paper tackles the problem of validating reinforcement learning agents under distribution shifts by generating counterfactual trajectories that show agent behavior in diverse unseen states, and demonstrates in a user study that their method enables better performance on one of two validation tasks compared to baselines.
In order for humans to confidently decide where to employ RL agents for real-world tasks, a human developer must validate that the agent will perform well at test-time. Some policy interpretability methods facilitate this by capturing the policy's decision making in a set of agent rollouts. However, even the most informative trajectories of training time behavior may give little insight into the agent's behavior out of distribution. In contrast, our method conveys how the agent performs under distribution shifts by showing the agent's behavior across a wider trajectory distribution. We generate these trajectories by guiding the agent to more diverse unseen states and showing the agent's behavior there. In a user study, we demonstrate that our method enables users to score better than baseline methods on one of two agent validation tasks.