Human-Like Motion of Robotic Arm Based on Dynamic Movement Primitives and Deep Neural Network

Human-like motion is necessary for successful human-robot collaboration. To achieve human-like motion with generalization capability on a robotic arm, we use dynamic movement primitives (DMPs) to model the motion of the human arm, i.e., DMPs extract parameters from the trajectories involving position and orientation of the hand, the elbow elevation angle. When using DMPs for human-like motion generation, i.e., planning human-like motion trajectories for robotic arms, the DMPs must be provided with targets. The task performed provides targets of end-effector position and orientation for the DMPs. We use a deep neural network (DNN) to provide the target of elbow elevation angle. The proposed method is validated in a simulation environment, and the Pearson correlation coefficient is used to evaluate the human-like level of the motion. The results indicate a high anthropomorphism of the robotic arm in performing generalized motions.