眼球機器人智慧型多軸追蹤控制

Abstract

本論文目的在於利用一個近似人類學習機制的類神經網路架構去控制四軸的眼球機器人，以追蹤移動的物體。整個追蹤策略分述如下，首先是預設眼球機器人眼睛和脖子與被追蹤物體位置之關係，其次是利用倒傳遞演算學習法則的離線學習（offline training），來訓練類神經網路以符合預設的對應關係。接著是利用影像即時擷取技術來求得欲追蹤物體的位置，當作已訓練完成神經網路的輸入訊號。最後利用類神經網路求得輸出訊號，作為眼球機器人各軸的速度修正參數。此外，為了能夠適應各種不同速度的追蹤，系統會根據物體速度來調整眼球機器人各軸轉動的速度。

This thesis proposes a novel design method for a quarto-axis eyeball robot, based on the neural network frame and close to the human being’s learning mechanism, to trace a moving object. The tracing strategy is classified into four parts. The first part is to preset the relationship between the axial speeds of the eyeball robot and the position of the moving object. Next, the offline training of the neural network based on back propagation learning rule according to the preset relationship. The third part is to retrieve the position of the moving object in real time as the input of the well trained neural network. In the fourth part, the output of the neural network is obtained as the reference data to modify the axial speeds of the eyeball robot. In addition, the eyeball robot is capable of adjusting its axial speeds in accordance with the object’s speed for different velocities tracing.