On Study Of Human Control Strategy In Robotic Manipulation

In recent decades, considerable attentions have been devoted to the research of service robots, especially the robotic hand within them. In this paper, we applied the Human Control Strategy method to control the motions of robots to be more human alike. Before that, we give an introduction to the kinematical analysis for robotic arms,and a robot mimic system. Through forward kinematics, the robot could compute the end-effector position with joint angles, and in turn, the joint angles could also be obtained through the inverse kinematics, with the end-effector position given. A human imitation system for robot is also proposed which can mimic the motions of the whole body in real time.Formally, Human Control Strategy consists of a demonstration phase, a learning phase, and a reproduction phase. The human tutor demonstrates the robot how to accomplish a specific task several times and then human movement strategies within demonstrations are extracted by a learning process. In the final stage, the robot armed with these strategies is examined to reproduce the same tasks independently without any human guidance. The learned strategies have the capacity of generalization and thus could be applied to unseen areas.Human Control Strategy can also be called the Programming by Demonstrations or imitation learning, more specifically for the case of modeling of motions. This paper presents a method based on the Extreme Learning Machine for robots to learn motions from human demonstrations. A motion is modeled as an autonomous dynamical system and sufficient conditions are derived to ensure the globally asymptotic stability at the target. We give a detailed theoretical analysis on the constraints regarding to input and output weights which yields a stable reproduction of demonstrations. We solve the corresponding optimization problem with nonlinear programming and evaluate it on both available data set and a real robot. Combined with the special neural structure of the Extreme Learning Machine, the results show that the human movement strategies within demonstrations can be generalized well and the reproductions always converge to the target points.