多軸力規開發與其在多足機器人上之應用

Abstract

本論文致力於內嵌於六足機器人之多軸力感測器的開發，並提出不同的校正方式使得校正方法能夠更為貼近真實的情況。藉由修改力規校正時的固定方式，使真實機器人與地面之間的作用力傳遞到力感測器的方式和校正方式相同，如此力感測器量測到的資訊即為馬達輸出扭力、圓弧形腳對於身體產生的平行於運動方向以及重力方向的驅動力。此外，利用商用力規與自製力規進行訊號校正，並利用標準一公斤重物在靜態放置以及動態旋轉下對馬達框架的施力比較，以評估自製力感測器所量測到的資訊是否更接近真實情況。 原本在Two-leg Model下生成的動態步態軌跡由於真實機器人落地時的初始狀態與模型中的不同，造成動態步態的成功率無法很高，但在六足機器人身上整合了力感測器之後，能夠量測到多軸力與力矩來感知到機器人在環境中的狀態，並回饋到系統以修正機器人的步態軌跡，使其能夠在下一步開始的初始狀態與模型更為接近，藉此減少機器人產生非模型中的落地情況像是機身撞到地面，使得機器人能跑出更像模擬中的動態步態且成功率提高。 此外，可利用卡爾曼濾波器將IMU與傾斜儀整合為身體姿態，再加入力感測器獲得的軸向力資訊建立平面模型，可計算出機器人在運動時腳與地面的接觸位置，並以實驗的方式證實其可行性。

The purpose of this thesis is to develop a multi-axis force sensor that can be mounted into a hexapod robot, and propose a calibration method that is closer to realistic situation and use the force information obtained to develop a feedback control method to make the dynamic motion more stable. By revising the fixing method of the force sensor, the path of the force passing from ground to the force sensors is now the same as the calibration method. Through the comparison between self-made force sensor and commercial force sensor, and using a standard one-kilogram object to conduct both static and rotating experiment the accuracy of the force sensor is verified. When using the trajectories generated by two-leg model proposed in previous research the successful rate of achieving stable dynamic gaits is relatively low due to the difference of the initial condition between the model and realistic situation. With force sensors integrated, the force and torque information is fed back to adjust the robot trajectory. The feedback control changes the initial condition of the next step so that it is closer to the condition calculated with the model, and thus reduce the unexpected behavior such as robot body hitting the ground. The information obtained from IMU and clinometer is sent into Kalman filter to establish the body posture in space, and using the posture along with the force signal obtained from force sensor the surface model is developed. The contact point between the robot legs and ground can thus be known exactly. Experiments are conducted to verify the result.