This project is exploring how expert human knowledge can be encoded to teach robots to perform manual tasks more efficiently and cost effectively. A low-cost sensor system consisting of a hand coordinate frame, a Kinect motion sensing gaming device, and a Shimmer wireless sensing platform has been developed that is able to effectively measure hand motions of a human while performing a manual demonstration task. Bird re-hang was chosen as the initial demonstration task. It was shown that the sensor system could be used to obtain demonstration data for designing dynamic movement primitives that are capable of discerning motion nuances to distinguish between a proficient practitioner and a novice. The method is currently being applied to the task of bird deboning, which relies heavily on human finesse.
Project Contact: Ai-Ping Hu