Architecture for explainable reinforcement learning

This invention describes how to make artificial intelligence systems, particularly those that learn by interacting with an environment, explain their decisions. It achieves this by integrating explanations directly into the AI's internal models, affecting how it perceives its situation, chooses actions, and evaluates rewards. The claims specifically cover an AI agent that uses sensor data to estimate an "explainable reward" and then uses these explanations to manage a controller, with the explanations detailing the agent's actions or decisions.