Basic Research on Visual Servo System which Dynamically Combines the Position/Force Control and Environment Recognition

以往的视觉反馈控制方案假定关节伺服机构具有理想的跟踪性能，在设计视觉伺服系统时需要考虑机器人的动力学特性。因此，它们不适合高速运动。研究人员开发了一种新的视觉伺服方案，该方案将视觉伺服系统视为一个动力学系统，在任务级计算视觉伺服系统的逆动力学，但任务级逆动力学方案没有对象运动的模型。因此，它只保证了对象的步进运动的收敛，换言之，它是一类伺服系统。因此，如果物体连续运动，误差是不可避免的。研究人员制定了跟踪目标连续运动所需的条件，并提出了一种自适应估计目标运动的跟踪控制器，该控制方案由两个模块组成，一个是估计目标运动的非线性观测器，另一个是基于估计的目标速度的任务级逆动力学。研究人员证明了所提出方法的渐近稳定性。通过对两连杆直驱机器人的实时实验，验证了该方法的有效性。此外，在两连杆DD机器人上实现了位置/力控制器，并进行了球对球控制。球对球的控制类似于倒立摆。一个小球被控制以保持另一个大球的顶部，另一个大球连接到两连杆机器人的末端。利用摄像头获取小球位置，补偿力传感器的噪声输出。事实证明，视觉信息对稳定球是有效的。目前，力控制与环境识别的结合还不能令人满意，但本研究项目所开发的非线性观测器将在未来的研究中发挥重要作用。

Previous visual feedback control schemes assumed ideal tracking performance of the joint servo mechanisms.The robot dynamics wereneglected when designing the visual servo systems. Thus they were notsuitable for high speed motions. The investigators developed a newvisual servoing scheme which considers the visual servo system as a dynamical system and which computes the inverse dynamics of the visual servo system in the task level.However, the task level inverse dynamic scheme does not have the model of the object motion. Thus it only guarantees the convergence for the step motion of the object, in other words it is the type 1 servo system. Therefore errors are inevitable if the object moves continuously. The investigators formulated the conditions required to track the continuous object motions and proposed a tracking controller which adaptively estimates the object motions.The proposed control scheme is composed of two modules, one is a nonlinear observer which estimates the object motion and the other is the task level inverse dynamics based on the estimated object velocity. The investigators proved the asymptotic stability of the proposed method. The effectiveness of the proposed method is validated by the real time experiments on the 2 link direct drive robot. Moreover, a position/force controller was implemented on the 2 link DD robot and ball-on-ball control was performed. The ball-on ball controlis similar to the inverted pendulum. A small ball is controlled tokeep the top of the other larger ball which is attached to the end ofthe 2 link robot. The camera is used to obtain the small ball position which compensate the noisy output of the force sensor. The visual information is proven to be effective to stabilize the ball. At present the combination of the force control and the environment recognition is not satisfactory, but the nonlinear observer developed in this research project will play a important role in the future study.