A bayesian reinforcement learning approach for customizing human-robot interfaces