The Arm Planning with Dynamic Movement Primitive for Humanoid Service Robot

摘要

In order to realize the autonomous motor learning skills of humanoid service robot, we propose a systematic framework for trajectory planning and learning of robotic arms in this paper. The system is presented as a series of differential equations with good attractor properties called dynamic movement primitive (DMP). The core of DMP is to extract the motion characteristics from the human body demonstration, and then realize the reproduction and generalization of human actions by adjusting the nonlinear terms online. In this paper, a dynamic motion capture equipment is used to acquire the human's demonstration. The nonlinear terms of DMP is learned through local weighted regression (LWR) algorithm. Finally, the effectiveness of this method is verified by simulating the one-dimensional and two-dimensional trajectories in MATLAB and applying it to the existing household humanoid service robot in the laboratory.