利用學習法則於機器人之整體性精密度校正

Abstract

本論文提出一整體性精密度校正的方法來解決計算機輔助設計系統與機機 器人之座標等效問題。目前機器人精密度校正的研究大都著重於某些局部 區域，也就是說僅要求機器人在某些工作區域的誤差參數能達到所需的精 密度要求。這主要是由於不精確的誤差結果不能完全被模式化和限制辨別 誤差參數所需量測點的數目。為了去克服此僅是區域性準確的問題，我們 首先提出利用量測空間分析來合理分割工作區域成數個小區域和選擇在此 小區域的具代表性的誤差參數，並利用此有限的誤差參數經小腦模型算數 計數器或模糊小腦模型算數計數器神經網路的學習法則 來產生整個工作 區域的合理的誤差參數。最後，我們透過模擬及實驗來驗證此整體性精密 度校正方法的可行。 A global calibration scheme is proposed to resolve the coordinates' equivalence problem in integrating the CAD system and Robots. Current robot calibration schemes are inevitably with certain locality, i.e., the calibrated error parameters (CEP) will solicit the demanded accuracy only in certain region of the robot workspace. It s mainly due to that the errors resulting in the imprecision are not completely modeled and only limited number of measured data are available for identifying the CEPs. To tackle this locality problem, we propose first performing the measurement space analysis to appropriately divide the workspace into local regions and select the representative set of CEPs from each local region. Learning algorithms based on both the CMAC or FCMAC neural networks are then employed to generate appropriate sets of CEPs for the whole workspace based on the derived finite sets of CEPs above. Simulation and experiment are executed to verify the proposed global calibration scheme.