INTELLIGENT CONTROL OF TENDON-DRIVEN ROBOTIC MECHANISMS.

肌腱驱动机构可以独立于关节力矩调节机械关节刚度，对刺激扰动力具有鲁棒性。因此，它们适用于与力相关的任务，但由于非线性弹性，控制非常困难。这个项目有两个与这个特性相关的子项目。一个是开发肌腱驱动机器人机构的智能控制器，另一个是开发肌腱驱动的两足机器人，以吸收摆动腿与地面接触时的强烈冲击。对于智能控制，我们研究了(A)设计肌腱驱动机器人机构的稳定控制器和(B)给定任务的关节刚度优化。对于(A)，我们开发了由一个自适应模块和两个神经网络组成的人工神经网络控制器，并证明了期望力/位置轨迹的UBB。对于(B)，我们使用遗传算法对三个任务的关节刚度进行优化；击球，接球，想要的力量追踪和走球。我们开发了一个肌腱驱动的两足机器人。它几乎和一个高年级小学生一样大。每条腿有6个自由度，膝关节和两个踝关节由6根肌腱驱动。此外，新设计的非线性弹簧张紧装置（NST）安装在筋的末端。这使得半主动调节机械关节刚度成为可能，并保证了来自地面的脉冲反作用力的鲁棒性。另一个特点是膝关节处的关节扭矩分配不对称。也就是说，我们用三根肌腱来伸展膝盖，只用一根肌腱来弯曲它。结构设计使用了一款市面上可用的结构分析软件。设计了基于fpga的控制系统。采用遗传算法对行走关节刚度进行优化。仿真结果表明了该方案的有效性。

Tendon-driven mechanisms can be adjust the mechanical joint stiffness independently of the joint torques and robust for the stimulus disturbance forces. Therefore they are suitable for force-related tasks, while the control is very difficult due to the nonlinear elasticity. This project had two subprojects related to the feature. One of them is to develop intelligent controller for tendon-driven robotic mechanisms and the other is to develop a tendon-driven biped robot, to absorb the strong shock when the swing leg touches with the ground.For the intelligent control, we investigated (A) design stabilizing controllers for tendon-driven robotic mechanisms and (B) the optimization of the joint stiffness for given tasks. For (A), we developed ANN controllers composed of one adaptive module and two neural networks and proved the UBB of the desired force/position trajectories. For (B), we optimized the joint stiffness using GA for three tasks ; ball hitting, ball receiving, desired force tracking, and walking.We developed a tendon-driven biped robot. It has almost same size as a higher-class elementary school student. Each leg has 6 DOF and the knee joint and two ankle joints are driven with 6 tendons. Furthermore a newly designed nonlinear spring tensioning device (NST) is attached at the end of the tendons. This makes it possible to adjust the mechanical joint stiffness semi-actively and guarantees the robustness for the impulsive reaction forces from the ground. Other feature is the asymmetric joint torque allocation at the knee joint. Namely we used three tendons to expand the knee and only one tendon to bend it. The structure was designed using a commercial available structural analysis software. A FPGA-based control system was designed.GA was used to optimize the joint stiffness during walking. Simulation results showed this scheme would work well.