Asymmetric Reinforcement Learning

A training approach where the agent has access to additional information or supervision during learning that won't be available when deployed.