Planning of proactive behaviors for human-robot cooperative tasks under uncertainty

For seamless human–robot cooperation, a robot may need to take several steps proactively to minimize unnecessary delays between the human's intention and the robot's corresponding reactions. By predicting exogenous events from human intention and generating proactive plans based on the predicted events, a robot can reduce delays and significantly improve interaction. In this paper, we propose a decision-theoretic proactive planning framework that selects best proactive actions and the best times for those actions as a means to improving human–robot interactions. To this end, we developed a composite node temporal Bayesian network as an extension to handle both the nature of an event and its time of occurrence within a single framework. We also developed a composite node temporal influence diagram that combines a composite node temporal Bayesian network with a limited memory influence diagram to solve proactive planning problems. Experimental results obtained using a robot assistant in a manual assembly task demonstrate the effectiveness of our proposed framework.