Study on environment-adaptive manipulation system

机器人要在真实的世界中执行任务，需要对外界环境进行自我识别，并与未知环境进行主动接触。这些研究中使用的系统由视觉传感器和机械手、手等工作部件组成，视觉传感器的数量和布局根据目的而不同。视觉伺服系统有两种类型，“手眼系统”是将视觉传感器放置在机械手的末端，“手眼系统”是将视觉传感器与机械手分开放置。使用后一种系统，我们可以像人眼一样获得广阔的视野，并可以客观地看待工作环境，包括其终点。但是，在控制系统方面，由于是根据图像信息计算出的目标物的三维位置来控制机械手，因此容易受到摄像机的标定误差的影响，无法实现高的位置精度。本研究构建了具有接近人的结构的眼-手系统，能够在真实的世界中完成各种任务。通过与通用CCD摄像机组成的主动视觉系统协同控制机械手对运动目标进行跟踪，实现接触后，利用环境自适应力控制系统对目标进行抓取。提出了变刚度矩阵的柔顺控制系统来实现稳定抓取，并将该方法应用于机器人多指手的实验中，得到了以下结论.通过改变剪切方向力引起的关节刚度，得到了适应质量变化的抓取控制系统.在摩擦锥内设置适当的指力区域，通过将力矢量约束在该区域，实现了适当的力抓取。

Self-recognition of external environment and accommodative contact to unknown environment are required for robots to carry out tasks in real world. Systems used in these researches consist of vision sensors and working parts such as manipulators and hands, where the number and the layout of vision sensors vary depending on their goal.There are two types of visual servo systems; "eye-in-hand system" is the one with vision sensors placed at the end point of the manipulator, and the "eye-and-hand system" is the one with vision sensors placed separately from the manipulator. Using the latter system, we can obtain a broad outlook as with human eyes and can take an objective view of the work environment including its end point. In terms of control system, however, the manipulator is controlled based on 3D position of the target calculated using the image information, and therefore, the system is vulnerable to the effect of camera's calibration error and can't accomplish high position accuracy.In this research, we constructed the eye-and-hand system having configuration close to human for accomplishing various tasks in real world. The moving target is tracked by cooperatively controlling the manipulator with the active vision system consisting of general-purpose CCD camera, and after the contact is achieved, we aimed to grasp the target using the environment-adaptive force control system. Compliance control system with variable stiffness matrix is proposed to achieve stable grasp, and by handling the experiment using the multifingered robot hand, which this method is applied to, we obtained the following conclusions.1. By changing the joint stiffness due to the shear direction force, the grasp control system adaptable to mass change was attained.2. Area of appropriate finger force was set inside the friction cone, and by constraining the force vector in that region, grasp with appropriate force was achieved.