基於運動學之全自主機器人動態運動規劃系統

Abstract

當居家服務型機器人提供服務時，一般會先對環境進行感測，透過感測器的使用，偵測環境的情況再進行運動策略。早期文獻研究主要集中於靜態的環境，環境結構固定，無移動之物體，機器人藉由感測器感知之資料來進行自我定位與建立環境地圖，並規劃路徑，使機器人避免碰撞到達目的地。近年來，由於科技技術的演進，許多學者逐漸開始討論動態環境下機器人的移動規劃。動態環境本身具相當之挑戰性，感測器的效能、運算速度要求高，再者動態環境中移動物體之辨認、追蹤、碰撞預測與避障策略的設計，都相對在靜態環境下有較嚴苛的挑戰。 本篇論文透過運動學的方法來探討機器人與真實動態環境的互動關係，擬欲開發能依據不同的駕駛習慣來調變機器人與真實環境之間的互動模式。一般來說，避開障礙物主要透過幾種方式：一、維持定速，改變方向。藉由方向的調整避免產生碰撞。二、方向不變，改變速度，在原定的路徑上加減速，先一步通過預定路徑或待對方通過而避開碰撞。三、同時改變方向與速度。此一方式較上述兩者較為高級也相對於複雜。於此，本篇論文旨在探討機器人在維持原速度的條件下，根據環境中的運動模型預測，提出尋找免於碰撞的中繼姿態(via-postures)的方法。利用這些中繼姿態產生數條與原路徑接合之滑順路徑，最後經由不同權重函數的參數分析比較，遴選出符合期待的路徑，使機器人在移動過程中避開碰撞，完成任務，抵達目的地。 藉由MATLAB軟體工具的使用，進一步印證我們所提出演算法之合理性與適用性，同時也對不同權重參數所產生之效果進行分析比較，使該演算法能根據不同參數的調變，由函數權重的設置挑選出符合預期之路徑。於此同時，亦設計情境，結合實際硬體的使用，進行實作展示。經由理論的推導到軟體的演算法開發與實務硬體之操作，最後於結論中歸納出研究之成果與討論且提出未來待開發及發展之處。

These years, the applications of robot have become more and more prosperous. Robot become one part in our lives gradually, they may be companions in the future. This thesis attempts to handle the problems of the obstacle avoidance in dynamic environment. In indoor environment, there always exist moving objects those may be people or other service robots. For indoor service robot, it is fundamental and essential to achieve the destination without collision. In the thesis, we develop the collision-free algorithm by using the mobile robot kinematics. From the point of view, the relative relations are formulated and the collision is predicted. Here, we propose the collision-free conditions to find out proper via-postures. While the condition is satisfied, the relative distance between the robot and the moving obstacle will be guaranteed so that the collision also disappears and via-postures are derived from the condition. To plan a new and smooth path to substitute the original one, a novel path planner is introduced. A smooth path is generated with two given postures. By using the path planner, the robot redirect to the original route to continue to move. Simulation of the algorithm is made to exam the properties of the algorithm. Simulation and experiment are both conducted to verify the approach. In the scenario, the robot follows the route and there are a man and a robot move straightly across the route. Robot will detour a new path if collision is detected and be redirected to the previous route after the risk of the collision vanishes. About the experiment, a scenario is designed to demonstrate, the webcam is set overhead to capture the information of the robot and the moving objects. All the information is computed on the host server to do the prediction. And the host server sends the commands to control the robot to be collision-free. The scenario demonstrations are conducted to extend the simulations into reality. It demonstrates the applicability of the approach. Finally, conclusions come out and the future works are listed to improve and develop the approach.