基於控制障避函數之人形機器人全身控制架構

Abstract

在機器人領域中，將具有高自由度（DoFs）的人形機器人應用於現實世界之各個場景仍然是件艱鉅的挑戰。過往的研究通常是採用條件與假設的簡化，如此雖然利於基礎的應用，但顯著地限制機器人的靈活性和適應性。為了應對現實世界的多樣性以及複雜條件，機器人必須採用不同的調整策略；不僅如此，為了保障應用的可行性，機器人在操作上的安全議題尤為必要。上述現象更加凸顯出擁有一個通用且靈活的控制框架對於現有人形機器人的操作效能與安全至關重要，亦是使其順利融入人類社會不可或缺的要素。 本文旨在提高仿生人形機器人對於人類社會之適應能力，並從中取得重大進展。主要有七項貢獻，這些貢獻進而增強了機器人的設計與功能性。首項創新是浮體運動學，使機器人得以將其基座設置在任意位置，提高了工作空間和穩定性。第二件是質心動量矩陣（Centroidal Momentum Matrix, CMM）的遞迴演算法，得以高效計算出每一機構之間的動量，該項目於運動規劃中扮演無可替代的角色。第三個為一種加速蒙特卡羅 (Monte Carlo) 分析的演算法，在降低複雜性的同時更大大減少運算時間。第四項則是機器人自碰撞避免之概念的引入，使用邊界球體和控制障避函數（Control Barrier Functions, CBFs）得以確保機器人整個運動過程之安全性。第五項貢獻乃基於CBF-二次規劃（Quadratic Programming, QP）的控制架構，並使用比例－微分（Proportional-Derivative , PD）控制法則以實現穩定行走。貢獻六則是將運動任務分類為全局任務與基礎任務，進而簡化軌跡之設計規劃和動作之復原校正。最後，本文提出一種角動量變化率之補償方法，使用力－扭矩感測器的回授，達成機器人之動態穩定。 這些貢獻提升機器人的效率、功能性和適應性，同時大大提高其融入人類社會的潛力。

The challenge of deploying humanoid robots with high degrees of freedom (DoFs) into real-world scenarios remains a pressing issue in the robotics community. Previous works in this field have often relied on simplified assumptions, which, while useful for basic applications, significantly limit the flexibility and adaptability of the robots. To address the diverse and complex conditions of real-world use, robots must be equipped to adopt different strategies. Additionally, concerns such as the safety of these robots cannot be overlooked. These factors underscore the need for a more general and flexible control framework, which is crucial for the operational efficacy and safety of existing humanoid robots. This thesis presents significant advancements in the field of humanoid robotics, moving beyond these traditional limitations. It introduces innovations that enhance robot design and functionality in various ways. One such advancement is the development of floating-based kinematics, which allows robots to set their base at any arbitrary point, thereby expanding their workspace and enhancing stability. Another is the introduction of a recursive algorithm for the centroidal momentum matrix (CMM), crucial for efficient motion planning. Further advancements include an algorithm that accelerates Monte Carlo analysis, significantly reducing complexity and saving time. The thesis also proposes a novel self-collision avoidance algorithm that employs boundary spheres and control barrier functions (CBFs) to ensure safety during operations. A notable contribution is the development of a CBF-based quadratic programming (QP) based control framework that integrates a proportional-derivative (PD) control law, ensuring stable walking. The categorization of motion tasks into global and base tasks simplifies trajectory design and motion recovery, while the proposed angular momentum rate change compensation method, using force-torque sensor feedback, enhances dynamic stability. Collectively, these advancements mark a significant leap in humanoid robotics, improving the robots'' efficiency, functionality, and adaptability, and paving the way for their integration into human society.