Machine Learning Driven Human Skill Transferring for Control of Anthropomorphic Manipulators

The kinematic mapping between human arm motions and anthropomorphic manipulators are introduced to transfer human skill and to accomplish human-like behavior for control of anthropomorphic manipulators. The availability of big data and machine learning facilitates imitation learning for anthropomorphic robot control. In this paper, a machine learning-driven human skill transferring for control of anthropomorphic manipulators is proposed. The proposed deep convolutional neural network (DCNN) model utilizes a swivel motion reconstruction approach to imitate human-like behavior for fast and efficient learning. Finally, the trained neural network is translated to manage the redundancy optimization control of anthropomorphic robot manipulators. This approach also holds for other redundant robots with anthropomorphic kinematic structure.