Policy Observability

New RL method uses agent actions to reveal internal state

Researchers have developed a method using reinforcement learning to train autonomous agents whose actions can reveal their internal state, even with communication limitations. This technique, termed Policy Observability…