應用於眼球機器人之智慧型多軸追蹤控制

Abstract

為了模仿人類雙眼視覺的追蹤，本篇論文針對目標物的追蹤控制設計了結合類神經網路的雙眼機器人。整個追蹤控制的策略可分為四個部分，首先是由雙眼機器人利用攝影機擷取移動目標的影像，並且判別它的位置所在。其次是設計目標位置中心和追蹤速度的關係當作脖子和眼睛的兩種訓練模式。接著再利用倒傳遞演算學習法則的離線學習（off-line training）來設計四種類神經網路，學習訓練模式所描述的目標位置中心對應速度之間的關係。最後，雙眼機器人採用了最佳的訓練結果做為它的控制器來追蹤移動的目標。這些智慧型的追蹤控制器經由實驗證明確實可成功地完成物體的定點追蹤控制和水平追蹤控制。

To emulate the humanoid binocular vision tracking, this thesis presents the tracking controller design based on the neural network for the Eye-robot to trace a moving object. The tracking control strategy is partitioned into four parts. First, the Eye-robot retrieves the image of the moving object from two cameras and identifies its position. Second, two training patterns concerning the relationship between the tracking velocity and object position are designed for the neck and two eyes to learn. Third, under off-line training four neural networks are constructed and trained by the back-propagation algorithm to learn the training patterns. Finally, the Eye-robot adopts the well trained neural networks as its controller to trace the moving object. The success of the tracking control of the Eye-robot can be concluded from the experiment results of the set-point control and the horizontal trajectory tracking control.