基於五質心及角動量模型之類人形機器人行走步態產生系統之研究

Abstract

為了讓雙足機器人或人形機器人能穩定地行走，零力矩點的理論被廣泛的應用，若零力矩點是在機器人行走時落於腳掌的安全範圍裡，則可判斷是穩定地行走。所以可以利用零力矩點理論來建立機器人的動態模型，倒單擺模型是最常用的模型，但此模型將機器人視為單質心，故有許多模型誤差。為了改善此模型誤差，後有飛輪模型考量此質心轉動慣量的影響因素。在雙足機器人裡，為了提高行走效率及減少模型誤差，有學者提出了三質心模型，將機器人視為三質心，分別為軀幹、左腳及右腳。為了提高三質心模型的精準度，我們考量各質心所產生的角動量，故我們提出了三質心及角動量模型應用於我們實驗室的第一代行走機器人-雙足機器人。後來我們實驗室製作出第二代行走機器人-人形機器人，若三質心及角動量模型應用於人形機器人，其人形機器人行走時的零力矩點誤差偏高。所以我們將人形機器人的雙手考量進模型，並提出了五質心及角動量模型應用於人形機器人，進而減少零力矩點誤差並提升人形機器人的行走效能。 本篇論文研究計畫所討論的步態產生系統有二種。第一、我們提出一個基於三質心及角動量模型使用模型預測控制之雙足機器人行走步態產生系統，此研究目標是降低模型誤差及減少預看的時間，進而提高零力矩點的追蹤及步態產生器的即時性。此研究貢獻是使用三質心及角動量模型來降低模型誤差，進而提升雙足機器人的行走效率與零力矩點的追踨精準度。此步態系統具有及時性，所以若機器人行走時遇到一些意外時，可立即產生新的行走軌跡來處理此意外。此步態產生系統有進行模擬且實現行走實驗於我們實驗室所開發的雙足機器人。 第二、我們提出一個基於五質心及角動量模型使用前饋與回饋控制之人形機器人行走步態產生系統，此研究目標是降低模型誤差及非最小相位系統的屬性，進而提升行走效率及零力矩點的誤差追蹤。此研究所提出的五質心及角動量模型因考量手臂及腳的轉動慣量，所以可以減少模型誤差進而提高人形機器人的行走效能。在非最小相位系統裡，因為頻率特性的關係，其極零點對消法是使用級數近似法，故此步態系統可以克服零力矩點理想軌跡的突然更改。此步態產生系統有進行模擬且實現行走實驗於我們實驗室所開發的人形機器人。

For the stable walking of the biped robot or the humanoid robot, the theory of zero moment point is widely utilized as a stability index. According to the zero moment point theory, the robot will be stable if ZMP moves in of the support polygon. The walking robot is supposed to be modeled dynamically based on zero moment point theory, so that linear inverted pendulum model is the simplest model. Because the model views all mass is concentrated at a point named the center of mass, the modeling is produced. In order to improve the modeling error, the flywheel model takes the inertia of center of mass into account. A researcher proposed three-mass model to enhance the zero moment performance, the walking robot is considered as three points including trunk, left leg and right leg. In order increase the accuracy of the model, we propose three-mass with angular momentum model which is applied in the first generation walking robot, biped robot, in our lab. After we develop the second generation walking robot, humanoid robot, the tracking error of the zero moment point is increased if three-mass with angular momentum model is applied in humanoid robot. Therefore, considering two arms into model, we propose five-mass with angular momentum model for humanoid robot to decrease the tracking error of the zero moment point and increase the walking performance of the humanoid robot. There are two approaches of the walking trajectory generation in this dissertation, including biped walking trajectory generation using model predictive control and quasi-natural humanoid walking trajectory generation using feedback-feedforward control. Firstly, we present a biped walking trajectory generator based on three-mass with angular momentum model using model predictive control. This approach aims to decrease the modeling error and decrease zero moment point horizon, so that high zero moment point tracking accuracy and immediate generation are achieved. The contribution of this approach is use of three-mass with angular momentum model, the modeling error is reduced and zero moment point tracking performance and walking stability are enhanced. This method allows online walking pattern modification, so that unexpected emergency can be dealt with by immediately changing trajectories. In addition, this proposed biped walking trajectory generator is validated through numerical simulations and the proof-of-concept experiments are conducted using experimental biped robot developed in our laboratory. Secondly, this dissertation proposes to develop a quasi-natural humanoid robot walking trajectory generator based on five-mass with angular momentum model using feedback-feedforward control. This approach aims to minimize modeling error and improve the frequency characteristics from non-minimum phase properties so that walking performance and tracking accuracy are enhanced. This proposed model focuses on the angular momentum effects from arm and leg rotation to reduce modeling error to enhance walking performance. Based on pole-zero cancellation using series approximation method, it can overcome the sudden change of the natural zero moment point reference due to the frequency characteristics in the non-minimum phase control system. The humanoid walking pattern generator is verified and demonstrated using a humanoid robot developed in our laboratory based on five-mass with angular momentum model.