Modeling and Model Learning in Very High Speed Visual Servo Systems

点击翻译按钮获取中文摘要

Conventional manipulators are designed to control the end point position with high accuracy based on the joint angle measurement, and are not designed for high-speed control with vision sensors. Therefore, it is difficult to achieve an intellectual, high-speed task for enabling the performance of high-speed vision. And thus the speed-up of the whole visual feedback system highly depends on the speed-up of the manipulation system. We think that introducing high-speed actuator is not enough and it is important to understand the characteristics of the image processing system and the active vision, the hand system and also the design of the control system. In this sense, to achieve the super-high-speed visual servo system, we have to develop the model that describes the dynamics of all elements composing the system ; and the development of the parameter identification technique ; and the model learning algorithm. The following research are done this year :1.Visual servo by high-speed hand system : A high-speed multi finger hand system was developed. An algorithm to catch a falling ball and a falling column stably was proposed, and the performance of the super-high-speed visual servo was evaluated by experiments.2.Microorganism tracking system : A system that can stably observe a swimming paramecium was developed. The position of the pool where a lot of paramecia swim was controlled by a XY stage.3.Expansion of stability area of visual servo : In this research, the method of expanding the stability area was examined in consideration of the restriction with limited camera view.4.Composition and robustness in visual servo system : In this research, it was pointed out that the robustness of the system changes greatly by combining different structures and algorithms.5.Ball capture and lifting task : Ball catching and lifting experiments have been done.