- 學位論文
全向機器人以演化式適應預測導引的大型延遲系統進行立體視覺伺服自主入庫
Adaptive prediction stereo-vision-servo design of large-scale omnidirectional robot for entering garage
摘要
本研究是假設機器人上沒有搭載攝影機，而且機器人欲駛入的車庫上面有遮蔽物(車庫頂)，因為假設影像死角及車庫頂太低的關係，所以攝影機不適合架設在過去文獻上所說的天花板、車庫頂或車庫周圍處，因此本研究是將兩顆無線攝影機架設在車庫入口正前方的某處合適位置，以完全不阻礙全向輪型機器人的行進為原則來架設這兩顆無線攝影機，並使用立體視覺技術及LAB顏色空間的強健濾波方法以全向輪型機器人上的兩個倒車燈的特徵點空間座標來決定機器人的現在姿態。因為影像處理所造成的時間延遲效應的問題嚴重，我們稱之為視覺迴授的延遲(Visual delays)，因此我們設計一套基於NARMAX控制器及類神經網路的受控場模型所提出的適應預測控制策略來減少時間延遲對系統控制的不良影響。NARMAX類神經網路控制器需經由演化(利用INN演算法最佳化控制參數)轉換成自我學習的人工智慧控制器，來進行影像伺服控制使得機器人自主駛入車庫。這個研究的目標是以立體視覺伺服來完成全向輪型機器人自我入庫實驗。
並列摘要
This research study develops a vision-servo three-wheeled omnidirectional robot to serve the human by using adaptive prediction control system and related software’s and hardware’s tool. In this research, we consider the visual-delays control problem of developing an application or experiment of robot without any camera for entering garage itself by using the two cameras mounted on the far side of this garage. Note that we don’t place any camera on this robot in this study. Turn on the two lights which are mounted on this robot when autonomous navigation starts, these positions of two lights on this robot are measured and estimated by using LAB (LAB is a kind of color space) filter, stereo-vision method and the neural network to obtain the stereo-vision-based information of the future pose of this robot. Moreover, this stereo-vision-based information is applied to the input of the evolutionary NARMAX adaptive predictive controller and the optimization of control parameters of this neural controller by using the proposed new genetic algorithm, INN(Island-type Neural-Network method), for this robot’s mission of entering garage itself. The purpose of this study is to finish the mission of entering its garage of three-wheeled omnidirectional robot by using stereo-vision technique.