Human training using HRI approach based on Fuzzy ARTMap networks

Based on recent studies which establishes that skill acquisition requires not just specification of motor skills, learning and skill application but also intervention of human expert only in certain phases, we present an approach which encode the human expert demonstration into a teacher class based on Fuzzy ArtMap network. Then, the human novice trainee produces the approximate knowledge, which is in turn coded into student class. The evaluation function introduces a class metric which simultaneously allows the student to refine motor commands to increase the trainee pace while modifies accordingly the desired trajectory of the robot. Preliminary experiments indicates a high success rate in contact robotic tasks, in a deterministic setting.