機器人學習控制之學習空間範圍與分析

Abstract

為了處理機器人動態的非線性，已經有許多學習控制器經由不同的控制架構應用在機器人運動掌控上。但是大部分的學習控制器在遇到新的運動軌跡時，都必須重新執行學習的程序。否則，學習控制器如模糊系統、類神經網路等將需要非常多的法則或神經原去學習這些新的運動軌跡，而這完全是因為學習空間太大的關係。有鑑於人類運動程式的概念，我們認為有效地整合已學習過的運動技巧來掌控還未練習過的運動，應該是降低學習空間過大的主要關鍵。在本篇論文中我們提出了一個新的學習控制策略來增大經由學習空間分析之學習空間的範圍。在此學習控制策略中包含了一個新的學習架構，其主要是由模糊系統與小腦模式控制器型式之類神經網路所購成。模糊系統是用來掌控一群未分類運動中的一些取樣運動，而小腦模式控制器型式之類神經網路則是用來整合模糊系統中之參數，以達成掌控整群運動的目的。同時我們也藉由量測運動間的相似性將整群運動加以分類，使得小腦模式控制器型式之類神經網路能夠更有效地進行整合。因此掌控大範圍機器人運動的學習難度將會被大幅降低，而學習程序僅需進行一遍。

To tackle the nonlinearity present in the dynamics of robot manipulators, the learning controllers have been applied for robot motion governing using different kinds of control structures. However, most of them need to repeat the learning process each time a new trajectory is encountered. Otherwise, a neural network will consist of a huge number of neurons or a fuzzy system will require too many rules because the learning space needed to handle arbitrary trajectories is too large. Inspired by the concept of human motor program, we consider that to generalize the learned motions effectively for governing those unpracticed motions is the key role to reduce the size of the learning space. In this dissertation, a novel robot learning control scheme is proposed to enlarge learning space coverage based on learning space analysis. A new learning control structure in the proposed scheme consists mainly of a fuzzy system and a cerebellar model articulation controller (CMAC)-type neural network. The fuzzy system is used for governing a number of sampled motions in a group of motions which are not yet classified. The CMAC-type neural network is then used to generalize the parameters of the fuzzy system, which are appropriate for the governing of the sampled motions, to deal with the whole group of motions. We also evaluate the similarity between robot motions so as to classify those motions governed by the fuzzy system, and make the subsequent generalization executed by the CMAC-type neural network more effective. Therefore, the learning effort is dramatically reduced in dealing with a wide range of robot motions, while the learning process is performed only once.