Ingmar Posner

The paper proposes a coherent inverse reinforcement learning (IRL) method to improve large behavior models for robotic control, achieving superior sample efficiency and performance on complex sparse manipulation tasks compared to traditional RL baselines.