人型機器人之最佳化步態生成與即時控制

Abstract

早從二十世紀初，人類對於人型機器人的幻想就不曾停止過，隨著時代的演進，我們對於人型機器人的想像一直都是類似的，會跑、會跳、行為舉止要像人類、甚至是超越人類，都是我們對於機器人的期待，可惜由於材料與致動器的限制，機器人的重量與力量輸出的比例一直沒有辦法追上人類，導致機器人的身體能力一直都沒有辦法突破到符合我們想像的程度，直到最近，由於科技的進步以及經驗的累積，才由日本與美國的研究者開發出具有跑跳能力並且十分穩定的人型機器人，機器人的身體能力也才慢慢開始足以應付這些困難的要求。但是在運動規劃方面，目前在學界與業界的研究上，仍然是各自採用各自的機器人平台與演算法來開發機器人的控制與運動規劃，各種的方法也有各自的限制與長處。有鑑於此，本論文提出了一套可通用於人型機器人之最佳化運動控制與步態生成器，此方法兼具了即時性與機器人重心高度與地面高度可變之特點，可達成自由指定零力矩點(Zero Moment Point)軌跡與重心高度軌跡輸入之步態生成。利用此一步態生成器與牛頓—尤拉動力學(Newton-Euler dynamics)，我們可以設計一個代價函數(cost function)並且求得重心高度軌跡對於此代價函數之導數，進而求得在指定零力矩點輸入軌跡之下最佳化之重心高度軌跡，達成本論文所提出之人型機器人之最佳化3D重心軌跡之步態生成器。除此之外，本論文也利用多個狀態機(state machine)與USB-to-CAN-Bus通訊網路建立一套人型機器人之即時控制系統，利用此系統，也驗證了我們所提出的即時步態生成系統的效能。

Since early twentieth century, human continues to imagine the future of humanoid robots. As the time going on, we always wish humanoid robots can run, jump, act like human, and even be better than human. Because of the limitations of material technology and actuators, the ratio of weight and power output of robots cannot reach the same level as human. Until now, due to the improvement of technology and the accumulated experiences, researchers in America and Japan start to demonstrate that their new robots can run and jump smoothly and stably. And the robots nowadays start to be capable of these difficult tasks. However, in the aspect of motion planning and pattern generation, the researchers in academy and in industry use their own robot platforms and algorithms to develop their motion control and planning systems. These systems have their own advantages and limitations. In view of this, an optimized walking pattern generator for humanoid robots is proposed in this dissertation. The proposed pattern generator can solve walking patterns with arbitrary assigned COG (Center of Gravity) height trajectory and 3D ZMP (Zero Moment Point) trajectory in real-time. Thus, walking pattern generation with arbitrary assigned ground height status is achievable. Based on the proposed walking pattern generator and Newton-Euler dynamics, a cost function is designed to optimize COG height trajectory with given ZMP trajectory. An optimized 3D COG walking pattern generator can be achieved in this dissertation. In addition, state machine based distributed control system with USB-to-CAN-bus interface is used to construct a real-time robot control system. Using this system, the performance of the proposed walking pattern generator is also verified.