Basic Research of Visual Servo which Dynamically Combines Environment Recognition to Motion Control

视觉伺服系统是一种基于图像信息的反馈控制机制，在非结构化环境中工作的智能机器人的运动控制中起着至关重要的作用。在传统的视觉反馈系统中，环境识别和运动控制是以逐步的方式顺序执行的。然而，这些传统的视觉和运动控制方案在快速、准确的操作方面存在不足，本研究的目的是将环境识别和运动控制动态地结合起来，开发一种识别和操作过程相辅相成的新的视觉伺服方案。我们提出了一种基于图像的动态视觉伺服方案，该方案比传统的视觉伺服方案具有更高的精度，并且对测量噪声具有较强的鲁棒性。该视觉伺服系统的动态特性由状态方程表示，基于状态方程的最优调节器可用于稳定系统，实现该视觉伺服系统需要相当强大的计算设备。因此，我们开发了一个由十个32位微处理器组成的并行处理系统--Transputers。在该硬件系统和PUMA 560上实现了基于图像的实时视觉伺服。本课题所提出的视觉伺服方案是将已有的控制理论应用到基于图像的视觉反馈系统中，并没有给出新的视觉反馈控制理论框架。我们发现，要建立新的框架，必须将机器人动力学包含在视觉伺服系统的动力学中，并且必须将包容机构视为增广的动力学系统。本课题的研究成果将成为视觉反馈控制理论未来发展的基础。

Visual servo system is a feedback control mechanism based on image information, which is essential in motion control of intelligent robots working in unstructured environments. In the conventional visual feedback system, environment recognition and motion control are performed sequentially in step by step manner. However, these conventional look and move schemes are deficient for fast and accurate manipulation.The purpose of this research project is to combine the environment recognition and the motion control dynamically so as to develop a new visual servo scheme in which the recognition and manipulation processes proceed complementing with each other. We have proposed a visual servo scheme called image-based dynamic visual servo, which is shown to be more accurate than the conventional scheme and to be robust against the measurement noises. The dynamics of this visual servo system is expressed by a state equation, based on which the optimal regulator can be used to stabilize the system.Realization of this visual servo system requires considerably powerful computing devices. Thus, we have developed a parallel processing system composed by ten 32-bits microprocessors--Transputers. Real-time image-based visual servo is implemented on this hardware system and PUMA 560. Though any Realistic manipulation can not be performed, some basic experiments verify the effectiveness of the proposed visual servo scheme.The visual servo scheme developed in this research project is derived by applying the available control theory to the image-based visual feedback system, which gives us no novel control theoretic framework of visual feedback. We found that, to build up new framework, robot dynamics must be included in the dynamics of visual servo system, and the inclusive mechanism must be treated as an augmented dynamic system. The results of this research project will become the bases of the future developments of visual feedback control theory.