Learning assistive strategies from a few user-robot interactions: Model-based reinforcement learning approach.