本研究整合高速影像系統與機器手臂之運動控制,執行擊球至目標點之任務。透過600FPS的高速雙眼視覺擷取球體三維座標,結合延伸式卡爾曼濾波器及球體飛行模型進行球體飛行軌跡的預測。以提高取樣頻率的方式讓球體飛行軌跡的估測可以更加精準,進而提升擊球機器人的打擊成功率。 機械手臂方面則是控制自行設計的四軸機械手臂(末端為拍子)進行擊球,並控制球體打擊後的落點。讓球體成功回擊至目標位置需要碰撞模型來描述球體碰撞瞬間的速度變化,準確的碰撞模型可以提高將球體回擊至目標位置的成功率。以往的研究是以球體與機械手臂拍面的物理碰撞關係進行討論,但是效果並不理想。因此本研究利用額外一組600FPS的高速雙眼視覺記錄球體與拍面碰撞瞬間的球體資訊,進而使用類神經網路對碰撞模型進行擬合。最後透過該類神經碰撞模型準確的描述碰撞瞬間球體與拍面的速度關係。
This thesis is dedicated to integrating a high-speed vision system with a robot motion control system to carry out ball-batting tasks which attempt to send the rebounding ball to the desired location. The 600FPS stereo vision system is used to acquire the position of the ball in the 3D space, and then extended Kalman filter and the ball flying model are applied to predict the ball flying trajectory. The high frame rate of the stereo vision system could reduce the error of the predicted trajectory and increase the success rate of hitting the flying ball. A robot with a batting pad at its end is designed to hit the ball and send the rebounding ball to the desired location. Therefore, an accurate rebounding model is required to describe the velocity relations among the incoming and rebounding ball trajectories and the pad. A physical rebounding model derived in the previous research is not accurate enough and results in low success rate for the batting task. Therefore, in this thesis, the bouncing trajectory is recorded by another high-speed stereo system and the data are used to train a neural network which learns an accurate relation among the velocities of the incoming and rebounding trajectories of the ball and the pad.