Interpretable Reinforcement Learning: Bridging the Gap between Performance and Transparency

Complex sequential decision-making tasks have proven remarkably successful for Reinforcement Learning (RL). However, the intrinsic complexity of RL algorithms frequently results in opaque, black-box models that are difficult to read and comprehend. In order to achieve a balance between performance and transparency, this study examines the changing environment of interpretable reinforcement learning. We introduce IRL and thoroughly assess five different IRL approaches, providing insights into each method's advantages and disadvantages. We evaluate RL and IRL's performance and transparency. Our results underline the revolutionary potential of IRL by highlighting its capacity to raise user confidence, responsibility, and safety by offering interpretable insights into RL agent judgement.