EARL: A Python Library for Counterfactual Explanations in Reinforcement Learning

A new Python library named EARL (Explanations using Alternative Realities for Reinforcement Learning) has been introduced to address the lack of transparency in reinforcement learning (RL) policies, particularly those based on deep neural networks. The library, described in a paper on arXiv (2608.14620), generates counterfactual explanations by exploring 'what-if' scenarios, allowing users to compare possible outcomes and better understand agent behavior. This approach is grounded in psychology research, which has shown counterfactual explanations to be intuitive and user-friendly, but its application in RL has been limited to toy examples and benchmarks. EARL aims to support counterfactual explanation generation in realistic RL-based self-adaptive systems, thereby enhancing user trust and facilitating system verification. The paper was announced as a cross-type submission on arXiv.