学习式运动控制在视觉引导机械手臂的应用

Application of Learning Motion Control for Visual Guided Mechanical Arm

针对一般的机械手臂运动规划时只适用于特定的工作任务,无法即时与环境互动的问题,开发出多轴学习式运动控制结合影像技术的机械手臂,以胜任各种操作任务的需求。通过建立矢量积分到端点(Vector Integration to Endpoint,VITE)模型实现即时轨迹规划,并采用小脑模型控制器(CMAC)作为学习式控制器,以应对较为复杂的工作环境,利用两台二轴旋转焦距缩放摄影机估测目标物位置,运用NI的嵌入式数字信号处理器和影像采集卡,进行影像的拾取与手臂的控制。实验表明,该机械手能够根据目标物的影像实现即时追踪抓取,其运动轨迹与PD控制器相比,其误差较小,且控制更精准与平顺。

General motion planning of mechanical arm is only suitable for specific tasks, and can not interact with the environment immediately. To solve this problem, a mechanical arm is developed by the combination of multiaxis learning motion control and image technology, which meets the needs of various operational tasks. First, the VITE (Vector Integration to Endpoint) model was established to realize real-time trajectory planning. Then, the cerebellar model controller (CMAC) was used as a learning controller to cope with a more complex working environment. Subsequently, the target location was estimated by using two axis rotating focal length zoom camera. Finally, the image collection and arm control were performed by using National Instrument (NI) embedded digital signal processor and image acquisition card. Experiments show that the manipulator can achieve real-time tracking and grasping according to the object image. The motion trajectory error of the developed manipulator is smaller than Proportion-Derivative (PD) controller, and its control is more precise and smooth.