Sample-Efficient Training of Robotic Guide Using Human Path Prediction Network

Training a robot that engages with people is challenging; it expensive to directly involve in the training process, which requires numerous data samples. This paper presents an alternative approach for resolving this problem. We propose human path prediction network (HPPN) generates user's future trajectory based on sequential actions and responses using recurrent-neural-network structure. Subsequently, evolution-strategy-based method only virtual movements generated HPPN presented. It demonstrated our proposed permits sample-efficient of robotic guide visually impaired people. By collecting 1.5 K episodes from real users, we were able train generate more than 100 required robot. The trained precisely guided blindfolded participants along target path. Furthermore, episodes, investigated new reward design prioritizes comfort during robot's guidance without incurring additional costs. expected be widely applicable robots interact physically humans.