Exploring the potentials of reinforcement learning in package delivery robot for improving human-robot interaction