Inverse RL via output feedback

Preprint. Model-based and off-policy inverse RL for zero-sum games from output data.

Model-Based and Off-Policy Data-Driven Inverse Reinforcement Learning for Linear Discrete-Time Two-Player Zero-Sum Games via Output Feedback. Preprint.

Quang Huy Dao, Quoc Dat Lai, and Phuong Nam Dao.

The learner imitates an expert zero-sum policy from output feedback, first with a model and then from input-output-disturbance data alone. The plots below are temporary result figures.

Output: expert and learner.
Equivalence of the recovered solution.
Disturbance attenuation.
Recovered solution.