Improved Bayesian inverse reinforcement learning based on demonstration and feedback