Development of a robot control system with a general task function

Abstract

Traditional control system is designed for specific movement. Whenever new tasks or motions need to be conducted, there will be a lot of additional work to change the control system. Meanwhile, human interaction with robotics become more common, and more complex tasks containing reaching and avoidance are needed. This project develops a new control system by applying machine learning method. A general task function with five parameters was introduced to describe the robot behavior. By learning the five parameters which related to potential energy during the training process, the robot will be able to follow and conduct similar tasks in the future. The project result will make the robot more user friendly and able to conduct more different tasks without explicitly programmed. Simulation and experiment results are presented to illustrate the performance of the new control system.