人形機器人之導航與軌跡規劃

Abstract

本論文提出一套自動導航系統可使機器人感知環境及其變化，並產生相對應的步態軌跡克服當下的環境變化。此方法結合環境感知(Environment Sensing)和步態軌跡規劃器(Footstep Planner)，其中環境感知包含了機器人本身所在位置、目標點的位置、以及環境中障礙物的位置及方向。而步態軌跡規劃器則是包含腳步轉換模型(Footstep Transition Model)、成本函數(Cost Function)、還有演算法(Algorithm)。此外，本論文更提出了針對腳步轉換模型的改善方法：動態腳步轉換模型(Dynamic Transition Model)。採用動態腳步轉換模型不僅可以增加規劃軌跡時的完整性(Completeness)，更可以增加規劃時的效率(Efficiency)。實驗環境包含靜態(Static)以及動態(Dynamic)來觀測機器人的自動導航系統的表現。本論文理論實現皆透過實驗室自主開發之人形機器人完成。

In this thesis, we present a navigation system that combines environment sensing with a footstep planner, allowing humanoid robots to autonomously navigate in unknown and cluttered environments. An environment map including the robot, goal, and obstacle locations is built in real-time from sensors. The footstep planner then computes an optimal sequence of footstep locations through footstep transition model, cost functions, and algorithm. Moreover, dynamic transition model is developed to improve system efficiency and completeness. The experiment results are used to demonstrate the robot navigating through both static and dynamic environments combined with sensors in real-time. The proposed dynamic transition model shows satisfactory results. All experiments are conducted on the NTU Humanoid robot, Nino, which is constructed by our laboratory.