In this paper we present a robotic system able to guide a person to a destination in a socially acceptable way. Our robot is able to estimate if the user is still actively following and react accordingly. This is achieved by stopping and waiting for the user or by changing the robot’s speed to adapt to his needs. We also investigate how the robot can influence a person’s behavior by changing its speed, to account for the urgency of the current task or for environmental stimulus, and by interacting with him when he stops following it. We base the planning model on Hierarchical Mixed Observability Markov Decision Processes to decompose the task in smaller subsets, simplifying the computation of a solution. Experimental results suggest the efficacy of our model.