基於五質心與角動量模型之人形機器人應用於多步態變化規劃與平衡

Abstract

在現代，機器人在我們生活中扮演著重要的角色，並且逐漸成為我們生活的一部分。在機器人的發展中，過去仰賴著舊式的輪型機器人，但其往往受限於環境的複雜性與工作的難易度，但隨著雙足機器人研究的興起、電腦科學的進步與硬體上的升級，人形機器人的出現與其具備之處理複雜工作、人機協力之能力，人形機器人在機器人領域中扮演著重要的角色。 為了使其可工作於這類環境，人形機器人的行走穩定性至關重要；在這領域中，我們常以零力矩點理論為判斷機器人是否處於穩定狀態的根據，如果零力矩點位於由腳底板畫出之安全範圍外，則判斷機器人可能會跌倒或不穩定平衡。 在過去，倒單擺模型提出將所有的質量集中於一質心點。而為了減少倒單擺模型之誤差，便有了更複雜的三質心模型和多質心模型的研究產生。除此之外，因忽略了轉動慣量與線性慣量之影響，倒單擺模型無法精確表示動態模型之系統，因此有將倒單擺模型改進之飛輪模型，其概念為將倒單擺模型的質心替換為一個有質量可旋轉的飛輪，以表示實驗上轉動慣量之影響。為了進一步提高模型的準確性與應用於人形機器人上，我們結合了三質心模型和角動量守恆系統，並提出了一個更精確更實用的五質心與角動量模型。該模型不僅考慮手部運動軌跡，也考慮了因兩腳行走所產生之轉動慣量對雙足行走的影響。換句話說，此模型以平衡腳部轉動慣量為出發點，以設計手部運動姿態。 因此，基於零力矩點理論，在本篇論文中提出一系列簡化多軸數系統之模型。為了驗證五質心與角動量模型之優點，我們採用了不同難度之步態規劃，其包含：踢腳、一般行走、突然變化行距之行走、曲線行走，和加入腰部軌跡之動作等。總結五質心與角動量模型之軌跡產生器，其特點在於結合，以五質心模型簡化多自由度之系統，並考慮過去質心模型多所忽略的轉動慣量對於動態系統之影響，最後以上半身(包含手部與軀幹)之運動來補償下半身所造成轉動慣量。

Robotics field have been a glowing research topics. Many conventional mobile robots are designed as wheeled type that are limited to the simple work and environment. But, as the mechanism and computer science are improved, humanoid robots are capable of handling more complicated works and co-work with human. To make sure the robot works well in that environment, walking stability plays an important role for this issue. The zero moment point (ZMP) theory has been proposed for determining whether robots fall down or not, i.e. stable or not. If the ZMP is outside of the footprint polygon, the bipedal robots might fall down or unstable. Linear inverted pendulum model (LIPM), regarded as the one-mass model, simplifies whole masses as a mass point. In order to reduce modeling errors LIPM has been implemented, and the three-mass model and multiple-mass inverted pendulum model (MMIPM) are developed. In addition to these researches, to reduce the error caused by the rational momentum and limited linear momentum, improved model with a flywheel showing embodiment of the centroid moment of inertia replaces single LIPM. To improve the accuracy of model and coherence to our mechanism, we further propose the five-mass with angular momentum model that combines the three-mass model and law of conservation of angular momentum. The proposed model not only considers the effect of rotational momentum, but also takes arm movements into account. In other words, as we design the leg motion, the balance of the moment of momentum through two arm motion is estimated in a meantime. Therefore, the research tends to simplify the high degrees-of-freedom actuator system to a simple model based on the ZMP theory mentioned in this thesis. To examine the performance of the proposed model, we further design different levels of motions. In this thesis, we have different motions, such as kicking, straight path walking, sudden changes of stride length, arbitrary curved path walking, and patterns with different waist height, etc. In summary, an effective walking pattern generator based on the five-mass with angular momentum model is proposed. The proposed model consists of two parts: the five-mass model simplified from a high degrees-of-freedom actuator system, and the balance of the rotational momentum while was not taken into consideration in previous researches. We utilize upper limbs, including arms and the trunk, to counterbalance the rotational momentum caused by lower limbs. The balanced motion is effective to reduce the modeling errors, and improves the ZMP tracking performance.