動作指令簡化之階段式學習方法-平衡點控制應用於軌跡追蹤之探討

Abstract

在本篇論文中, 經由觀察人類手寫字的學習過程而提出一個運動指令簡化 的階段式學習方法. 在我們的研究中認為, 類似簽名之類的熟練動作, 其 運動規畫應以速度為主. 本論文中所提出的方法, 包含了兩個學習階段, 可以將連續而複雜的運動指令簡化成簡單而不連續的運動指令. 在第一個 學習階段中, 連續指令是經由軌跡追蹤得到; 在第二個學習階段, 速度的 追蹤成為主要考量, 而指令的簡他則是漸漸地放棄速度追蹤的精確度. 我 們以平衡點假說的觀念控制動作的產生. 運動指令的簡化可以漸少學習控 制的記憶體需求. 人類簽名的實驗和模擬的結果驗證了所提方法之可行 性. 我們基於模擬的結果來探討平衡點控制應用於軌跡追蹤之可行性. In this thesis, a gradual learning scheme for motion command simplification is proposed inspired from the observation on human handwriting learning. Based on our study, we consider that skilled movement, such as signature generation, should be planned in the velocity domain. The proposed learning scheme contains a two-stage learning procedure, which can transform the continuous and complex motion commands into the simple, discrete commands. The continuousmotion command is derived at the first stage based on trajectory tracking. In the second stage, the velocity tracking dominates the movement generation and simplied motion commands are obtained by giving up the velocity tracking accuracy gradually. The motion control employed is based on the equilibrium point hypothesis. The simplification of motion commands will result in the reduction of memory requirement for a learning scheme. Experiments and simulations for human signature generation are performed to verify the effectiveness of proposed scheme. Based on the simulation results, the feasibility of equilibrium point control in trajectory is studied.