基於協同進化網路與決策樹之高效視覺校正系統應用於並聯式機器人

校準技術已被廣泛應用於機器人，校正為機器工具的準確度提供了具有成本效益的解決方案。標準的校正過程中，需進行三種不同的工作程序：建模，量測和校準。一般採用運動學模型建模，而對並聯式機器人來說是建立反向運動學模型。量測的方式一般採用高精度儀器對末端點做位置量測，收集一組完整的三維數據，建立系統的誤差模型。量測後就是校準，一般使用兩種不同的技術進行校準。第一種方法稱為“基於模型的方法”，要求在電動機與機械手的操作坐標系之間建立一個校準關係。這個關係需從機器人的幾何形狀、環境的誤差來源考慮近似的數學函數與係數參數。高精度多自由度的並聯機器人一般採用多項式模擬誤差模型，再對此誤差模型進行誤差補償與校準。第二種方法稱為“無模型的方法”，此方法中使用者並不需要知道影響機器人精度的主要誤差源，而是採用學習的方式去補償誤差。無模型的方法一般採用人工神經網絡來實現。以往神經網路的使用方式需要使用者反覆的調適網路架構，才有辦法得到好的學習結果。本論文提出讓使用者不需事先給定神經網路的架構，只需給定神經網路輸入與輸出，即可根據訓練的資料得出優化過後的網路架構，提高訓練效率並增加精度。與其他優化的神經網路比較，本論文提出的基於決策樹和協同網路的導傳遞算法得到更好的結果，不僅提高了方便性，也增加了準確性。此外我們提出的雙眼視覺系統輔以校正版，大大的減低了以往的雷射校正系統昂貴的負擔。這種技術對工業應用帶來廣泛的益處，利益在於該機器人在實際的微加工操作上透過軟體的技術去修正誤差，減少從硬體端修正誤差所帶來的龐大成本。

關鍵字

機器人校正；並聯式機器人；類神經網路；決策樹；協同進化網路

並列摘要

Calibration technique has been widely used in robotics, and it provides high accuracy and cost-effective solution for the robot. The calibration procedure only modifies the programming part instead of the hardware of robot design or tightening the manufacturing tolerances. This thesis proposed the highly accurate calibration method for the multi-DoF parallel robot. There are three steps in the standard calibration procedure: modeling, measurement, and correction. Kinematic modeling is a common way to modeling. Inverse kinematic model is used in parallel mechanism modeling. The high accuracy instrument is used to measure the position of end-effector. As for the position of end-effector, we can collect a complete set of 3D data and construct the error model. Correction starts after the measurement. The two different techniques are used in correction. The first method is model-based model. It constructs a correction relationship between the motor and operational coordinates of the robot. The correction relationship considers the geometric of robot and the environment error to decide the coefficients of mathematics function. The high accuracy multi-DoF parallel robot use polynomial function to construct the error model, and use this error model to compensate the error and calibration. The second method is model-free model. In this method, user doesn’t need to know the error source affecting the robot accuracy. The method compensates the error by learning method, and artificial neural network is commonly chosen for learning method. This thesis proposed an effective Visual Calibration System for Parallel Robot Using Cooperative Coevolutionary Network and Decision Tree Approach. The method can self-construct and optimize the neural network structure and parameters for the individual training set, and keeps the good prediction ability. This method combines with inverse kinematic model and it can find the accurate relationship between the motor and operational coordinates of the robot, and doesn’t need to consider the coefficient of polynomial and error model of the robot. The self-construction and optimization of the structure and parameters of neural network can help the robot achieve high accuracy in the workspace, and adapt in any source of error in the environment. The calibration technique brings a lot of benefit for the industrial applications, decrease the huge cost of error modification from the hardware and solves the problem using software technique.

並列關鍵字

robot calibration ； sub-mm accuracy ； parallel robots ； kinematic mechanism ； neural networks ； decision tree ； Cooperative Coevolutionary Network

參考文獻

[1] D. Stewart, “A platform with six degrees of freedom”, Proceedings of the IMechE, 1965.

[2] P. Maurine, E. Dombre, “A calibration procedure for the parallel robot Delta 4”, International Conference on Robotics and Automation, 1996.

[3] J.M. Perez, S. Schreiner, G.E. Gorton, “Evaluation of the VITUS Smart Laser Scanner for Accuracy, Resolution and Repeatability for Clinical Assessment of Pectus Deformities and Scoliosis”, Bioengineering Conference on Proceedings of the IEEE 32nd Annual Northeast, 2006.

[5] T.W. Wright, “Elements of Mechanics Including Kinematics, Kinetics and Statics”, London, E. and F.N. Spon, 1896.

[6] R.P. Paul, “Robot manipulators: mathematics, programming, and control: the computer control of robot manipulators”, The MIT Press, 1981.

國際替代計量

基於協同進化網路與決策樹之高效視覺校正系統應用於並聯式機器人

全文下載

主題瀏覽