Towards Learning to Handle Deviations Using User Preferences in a Human Robot Collaboration Scenario

Abstract

In a human robot collaboration scenario, where robot and human coordinate and cooperate to achieve a common task, the system could encounter with deviations. We propose an approach based on Interactive Reinforcement Learning that learns to handle deviations with the help of user interaction. The interactions with the user can be used to form the preferences of the user and help the robotic system to handle the deviations accordingly. Each user might have a different solution for the same deviation in the assembly process. The approach exploits the problem solving skills of each user and learns different solutions for deviations that could occur in an assembly process. The experimental evaluations show the ability of the robotic system to handle deviations in an assembly process, while taking different user preferences into consideration. In this way, the robotic system could both benefit from interaction with users by learning to handle deviations and operate in a fashion that is preferred by the user.