Experiments on collaborative learning with a robotic wheelchair

Abstract

To generate a path that guides the wheelchair's motion faces several challenges: The path is located in the human environment, which is usually unstructured and dynamic, and thus is difficult or impossible to generate a reliable map and plan paths on it by artificial intelligence. In addition, the path of a wheelchair, whose task is to carry the human user, should be smooth and comfortable, and adapted to the users intentions, which may evolve with time. We propose a collaborative learning strategy corresponding to these requirements, according to which the human operator and the robot, using the provided path design tools, create and gradually improve a guide path, eventually resulting in ergonomic motion guidance. This paper reports experiments performed to investigate this collaborative learning strategy. To evaluate the path design tools, we analyzed features of the optimal paths and user evaluation in representative conditions. This was complemented by a questionnaire filled out by the subjects after the experiments. The results demonstrate the effectiveness of the collaborative learning strategy, and show the utility and complementarity of the path design tools.