Vision based Neural Network Control of Robot Manipulators with Unknown Sensory Jacobian Matrix

摘要

Most research so far on task-space sensory feedback control of robot manipulators has assumed that the structure of kinematics or Jacobian matrix is known. As most industrial manipulators have closed architecture control systems and do not come with external sensors such as cameras, the sensors have to be added and integrated to the robots according to different requirements and applications. Since different configurations and types of sensors result in different sensory transformation or Jacobian matrices and thus lead to different models, it is in general difficult for operators or users in factory to model the sensory systems and deploy the robots according to various applications. Besides, as the sensory or visual feedback is implemented as an outer control loop in addition to the inner joint servo loop of the industrial robots, the interactions with the inner control loop must be carefully considered to ensure the stability of the overall system. This paper proposes a vision based neural network Jacobian tracking controller for robot manipulators. The proposed controller can be implemented on robots with either closed or opened architecture, without having to model the cameras, and manipulators. The effect of inner control loop is considered so as to ensure the stability of the whole system. The stability is shown by using the Lyapunov-like method and experimental results are presented to illustrate the performance of proposed controller.