雙輪式行動機器人之測距與定位特性分析及誤差消除

Abstract

在機器人自動導航中，含測距器的輪型行動機器人是相常泛用的平台之一。常見的使用方式，是利用一些理論或演算法來給予輪型行動機器人速度的指令，並接受其感測器回傳的資訊，以做為下一刻速度控制的參考值。而實際操作上，關於輪型行動機器人的感測層及致動層，都存在一些偏移誤差，以導致實驗時出現非預計的結果。 現今有許多研究，利用機率的模型或是強健演算法，以減少機器人的非偏移誤差，或是增加機器人對誤差的容忍度。然而，卻很少針對消除平均值不為零之偏移誤差的研究。 本篇論文採用雙輪式的行動機器人做為理論平台，並探討和機器人姿態相關的偏移誤差，也就是橫向打滑和縱向打滑。而為了修正機器人的姿態資訊，必須採用由外在感測器所建構成的里程計來量測橫向及縱向打滑。因為此兩種打滑的誤差，都屬於機器人與外界環境交互作用而來的誤差，是無法被傳統的車輪里程計所測得。 隨後，「里程計整合」和「誤差消除架構」這兩個重要的概念將被深入探討。前者是取得所有里程計的資訊之後，參考所處的環境狀況以及各類不同原理的里程計在不同環境的優缺點後，決定出一個最值得信賴的機器人姿態。後者除了包含了前者的概念，使機器人姿態能更為準確外，還會修正傳統車輪里程計的參數，使其之後測量到的結果，能直接屏除兩種打滑造成的誤差。

Wheeled Mobile Robots (WMRs) with range sensors are very popular platforms in robot navigation. Velocities of WMRs are controlled by some theories or algorithms, and sensor information of WMRs is received to decide new control commands. In real situation, some biased errors in actuating part and sensing part of WMRs will cause unexpected experimental results. Many studies have been introduced to correct unbiased errors by probability model and some robust algorithms are mentioned to let WMRs tolerate errors. However, few researches are mentioned about the bias error correction. In this thesis, a two-wheeled mobile robot (TWMR) is the platform and the bias errors about postures of TWMR, i.e., effects of skidding and slipping, are classified and modeling. In order to correct the posture information, the data of range sensors are processed as localization methods to eliminate the external error which encoder odometry cannot detected. After that, two concepts about localization method fusion and error elimination architecture are mentioned. The former combines the results of localization methods and determines the best results based on the performance of localization methods in different scenarios. The latter contains the former and makes the posture information more reliable. Besides, it corrects the encoder odometry with the knowledge of skidding and slipping.