Open Human-Robot Collaboration using Decentralized Inverse Reinforcement Learning