Inverse reinforcement learning for dexterous hand manipulation