Motion Control of Hyper Redundant Robots with Learning Control Scheme Based on Linear Combination of Error History

为了通过误差历史线性组合的学习控制建立有效的超冗余度机器人运动控制方案，讨论了设置合适的子任务和获得合适的机器人初始构形的新方法，基于机器人灵巧度的最优运动控制，以及带有弹性元件的机器人的柔性控制。研究结果总结如下：(1)针对超冗余度平面串联机械臂，提出了一种新的子任务设置法，以排除障碍物或辅助主任务。(2)提出了一种新的反向学习方案，将前向学习得到的收敛位形设置为初始位形，并重复学习过程，以获得合适的初始位形。(3)根据关节雅可比矩阵的列向量，推导出评价主任务可辅助性和杆件可移性的量化指标。以各指标为目标函数，采用梯度投影法和学习控制方法对关节输入进行优化。(4)提出了一种新的柔性控制方法，用于描述具有弹性被动关节的闭环冗余度机械臂的输出位移和刚度分布，并进行了实验验证。样机可以在控制输出柔度的同时操纵软/硬物体，这表明基于误差历史线性组合的学习控制可以在利用冗余性的情况下实现运动和输出力控制。

Aiming to establish effective motion control scheme for hyper redundant robot by learning control based on linear combination of error history, new methods to set suitable subtasks and to obtain suitable initial configuration of the robots, the optimum motion control based on dexterity of the robots, and flexibility control of the robots with elastic elements are discussed. The obtained results are summarized as follows :(1)For hyper redundant planar serial manipulators, new methods to set subtask so as to pass obstacles away or assist main task were proposed and formulated.(2) A new 'Backward learning scheme', in which a converged configuration obtained with forward learning was set as initial configuration and the learning processes were repeated, was proposed so as to obtain suitable initial configuration.(3)Quantitative indices to evaluate assistability for main task and movability of links were derived from column vector of Jacobian matrices of joints. By setting the indices as objective functions, the joint inputs were optimized with gradient projection method and learning control.(4) new flexibility control method to specify both of output displacement and stiffness distribution of a closed-loop redundant manipulator with elastic passive joints was formulated and was experimentally examined. A prototype can then manipulate soft/hard objects while controlling output flexibility.It was thus revealed that the learning control based on linear combination of error history could realize motion and output force control while utilizing redundancy.