雙足機器人步態之線上生成與多感測器回授控制

Abstract

本論文主要內容為雙足機器人具機動調控性之穩定步行步態的設計與開發，主要包含三個子項目：線上步態生成演算法、本體狀態(body state)估測器、以及多重感測器回授控制步態開發。在線上步態生成演算法方面，分成步態規劃與軌跡生成兩個階層，首先基於機構與安全的限制條件，規劃穩定步行路徑直到機器人到達預設的目標點，在步行過程中若預設目標點更動，路徑也會線上即時修正。接續以現階段規畫好的路徑為基礎，即時生成零力矩點與腳板軌跡，並由預觀控制（Preview control）法則生成質心軌跡，最後，由逆運動學求得各關節軌跡作為機器人底層的控制輸入。在質心姿態估測器方面，以擴展卡爾曼濾波器（Extended Kalman filter）為基礎，融合關節角度（編碼器）、慣性量測單元（加速規、陀螺儀）以及傾斜儀等感測器資訊，估測出機器人在三度空間中質心移動與本體轉動的完整狀態。在回授控制步態方面，藉由本論文提出的質心位置控制器以及引進的力矩控制器，即時調控機器人各關節驅動狀態以提高步行穩定性，其中質心位置控制器倚賴本體狀態估測器所提供質心追蹤誤差資訊，來調控機器人質心位置，而力矩控制器則調整踝關節角度以達成零力矩點分配器計算出的目標力矩。本論文並以實驗室現有質心加速度修補回授控制機制為基礎，以機器人簡化動態模型觀點出發，提出此控制器的參數設計流程。將此三個子項目整合，完成開發出一個具機動調控性之穩定步行步態所需的各個面向。最後，前述的三個子項目均已實際架設在實驗室現有雙足機器人上，並以實驗驗證其可行性與動態表現。

The main content of this thesis is design and development of a flexible stable walking gait for a biped robot, which contains three parts: online gait generation, body state estimator and multi-sensor feedback control gait. With respect to online gait generation, the algorithm is composed of gait planning and trajectory generation in layered structure. First, the gait planning algorithm designs stable walking gaits towards online destination point with constraint of mechanism and safety, and then the trajectory generation algorithm uses polynomial curve to generate Zero-Moment Point (ZMP) and foot trajectories according to planned gaits. After that, Preview control calculates trajectory of Center of Mass (CoM) corresponding to ZMP trajectory. Finally, inverse kinematics is applied to obtain the trajectories of each actuated joints in real-time, adopting specified values as control points. The proposed body state estimator is designed based on the Extended Kalman filter with sensory information from joint angles (encoders), an inertial measurement unit (accelerometer and gyroscope), and an inclinometer. The estimator estimates the full state of the CoM which contains translation and rotation motion in three-dimensional space. On the multi-sensor feedback control side, we propose a CoM position controller and introduce a torque controller to stabilize walking by modifying inverse kinematics in real-time. The CoM position controller relies on CoM tracking errors information from body state estimator to regulate the CoM of robot. The torque controller modifies joint angles of ankles to realize the foot reference torque calculated by the ZMP distributor, and this thesis proposes a design process to determine parameters of the controller based on existing CoM acceleration feedback control method from simplified dynamic model’s viewpoint. Integrating the three parts, we fulfilled the development of a flexible stable walking gait in many perspective. Finally, the former three parts is implemented on a child-size bipedal robot, and verified the feasibility and dynamic performance experimentally.