人形機器人即時行走軌跡規劃及最佳腰部軌跡生成

Abstract

雙足機器人相較於輪型機器人，在適應人類居住環境或克服天然不平地形有很大的優勢，而擁有兩隻腳的人型機器人則更能滿足為人類服務的需求。但在為人類服務之前，機器人必須有能力維持自身的穩定來行走在各種環境中，例如不平的地面，上下樓梯，或滿是障礙的空間等。至今已有非常多的理論被提出並使得機器人可以穩定的行走，預看控制就是其中非常重要的一項里程碑。基於先前研究人員提出的簡化機器人模型，例如單質心模型或三質心模型，在事先規劃好的腳步之下，我們可以利用預看控制來達到能讓機器人穩定行走的各軸角度軌跡規劃。 在本論文中，我們提出兩個可以增進機器人行走能力的全身性行走軌跡規劃演算法，其一是腰部最佳軌跡規劃演算法以及上半身動作補償，透過腰部數學模型的建立以及簡化，我們可以利用線性規劃解算器來求得最佳化的腰部軌跡，其效果能使得機器人夠擁有更快的走路速度以及更好運動能力，跨出原本無法達到的更大步伐。其二是線上走路軌跡規劃演算法，相較於先前必須先做好腳步規劃的離線版本，我們可以選擇使用人為控制或是機器人自動辨識的方式，來即時改變機器人的行走方向，結合近年的機器人行走理論，規劃出適當行走軌跡。最後，我們透過圖表來比較加上腰部運動軌跡之後的走路實驗，其行走表現以及運動能力均得到了提升，而我們所設計的即時行走軌跡規劃也確實能提供穩定且較為快速的走路運動。

Comparing with wheeled robots, biped robots have more flexibility of adapting to living environment of humans and confronting with uneven terrain. And biped robots with upper body and two arms, or humanoid robots, are more capable of meeting humans’ demand, for example, cleaning rooms or serve drinks. However, before starting to serve humans, the humanoid robot itself need to guarantee its stability while walking in all kinds of environment, such as uneven terrain, stairs and spaces occupied by obstacles. So far, there have been many successfully developed theories and implementation of walking pattern generator that can perform stable walking for humanoid robot, and preview control theory is one of the most important one in the field. Based on simplified model for humanoid robots with complex structures, linear inverted pendulum model for example, we can do the footstep planning in advance and utilize the preview control theory to generate walking patterns with whole-body motions. In this thesis, we propose two trajectory generation algorithms that can greatly enhance the capability of a walking robot. One of the algorithms is the waist trajectory generation algorithm that can generate optimal waist motions and upper body compensation motions. By formulating and simplifying the waist trajectory generation problem, we are able to utilize the state-of-the-art linear programming solver to generate optimal waist motions. The generated optimal waist motions then increase the walking speed and exercising capability of humanoid robots, lengthen the walking stride for example. The second algorithm we proposed in this thesis is the online walking pattern generator. The difference of the proposed online walking pattern generator and the previous offline walking pattern generator in our lab is that the robot’s walking direction can now be changed by humans or the robot itself in any time. The algorithm exploits the concept of capture points and plans proper footsteps and walking patterns. Finally, the experiments show that two proposed trajectory generation algorithms do increase walking speed, lengthen walking strides and still preserve stability.