基於人形機器人視覺之改良型蒙地卡羅定位法

Improved Monte-Carlo Localization Base on Humanoid Robot Vision

王亮欽(Liang-Chin Wang)

指導教授 ： 劉智誠

淡江大學/工學院/電機工程學系機器人工程碩士班/碩士(2020年)

https://doi.org/10.6846/TKU.2020.00787

摘要

本論文針對視覺自主之小型人形機器人提出一個在RoboCup足球場上基於人形機器人視覺之定位方法。在Linux環境下，以機器人作業系統(Robot Operating System, ROS)建構人形機器人的定位架構，在視覺方面，本論文利用邊緣偵測與拉普拉斯轉換(Laplace transform)進行影像前處理，接著透過掃描線來找出場地線上的特徵點，並利用逆透視映射法(Inverse Perspective Mapping, IPM)計算出特徵點與機器人間的距離，以此作為定位所需的觀測資訊。在定位系統中，本論文使用蒙地卡羅定位(Monte-Carlo Localization, MCL)作為其主要架構，並改進其三大問題：(1)粒子數量的調整、(2)陷入區域最佳解、(3)機器人綁架。粒子數量的多寡會影響蒙地卡羅定位的速度，本論文利用KL散度(Kullback-Leibler Divergence, KLD)根據定位對環境的相似度來調整粒子數量；為了解決區域最佳解問題，本論文使用競爭選取法(Tournament Selection)作為蒙地卡羅定位重新取樣的方法，該採樣方式可有效使蒙地卡羅定位在搜尋位置時保持粒子的多樣性；機器人綁架為機器人定位常見的問題，本論文採用Augmented-MCL中的方法來解決此問題，藉由加入滑動平均(moving average)來解決機器人綁架問題。綜合以上三點，本論文提出一種改良型蒙地卡羅定位法來解決以上三種問題。從實驗結果可知，改良型蒙地卡羅定位可以有效地解決以上三種問題。

關鍵字

視覺定位 ； 蒙地卡羅定位 ； 粒子濾波器 ； 競爭選取法 ； KL散度 ； 機器人綁架 ； 機器人作業系統

並列摘要

In this thesis, a localization system is proposed to implement on RoboCup soccer field for vision-based autonomous small-sized humanoid robot. In the Linux environment, Robot Operating System (ROS) is used to establish the localization system for the humanoid robot system. In visual system, edge detection and Laplace transform are used for image preprocessing, and Inverse Perspective Mapping (IPM) is used to calculate the distance between the feature point and robot. In localization system, Monte-Carlo Localization (MCL) is used for the main algorithm, but there are three problems need to improve in MCL: (1) the adjustment of the number of particles, (2) the local optimal solution, and (3) the robot kidnapped. Kullback-Leibler Divergence (KLD) is used to adjust the number of particles according to the similarity of localization and environment. In order to solve the problem of local optimal solution, Tournament Selection is used as the method of resampling in MCL. In order to solve robot kidnapped, the method in Augmented-MCL is used to solve the robot kidnapped by adding moving average. Base on the above three points, Improved Monte-Carlo Localization (IMCL) is proposed to solve the three problems in this thesis. In the experimental results, IMCL can effectively solve the three problems.

並列關鍵字

Vision-based Localization ； Monte-Carlo Localization ； Particle Filter ； Tournament Selection ； Kullback-Leibler Divergence ； Kidnapped Robot ； Robot Operating System

參考文獻

[1]R. E. Kalman, “A new approach to linear filtering and prediction problems,” Trans. of the ASM-Journal of basic engineering, no. 82, pp. 34-45, Mar. 1960.

[2]D. Fox, W. Burgard, and S. Thrun, “Markov localization for mobile robots in dynamic environments,” Journal of Artificial Intelligence Research, vol. 11, pp. 391-427, 1999.

[3]F. Dellaert, D. Fox, W. Burgard, and S. Thrun, “Monte Carlo localization for mobile robots,” Proceedings 1999 IEEE International Conference on Robotics and Automation, Detroit, MI, USA, vol. 2, pp. 1322-1328, 10-15 May 1999.

[4]S. J. Russell and P. Norvig, “Probabilistic reasoning over time,” Artificial Intelligence: A Modern Approach, Upper Saddle River, vol. 15, pp. 566-609, 2003.

[5]J. K. David, T. Ernst, and O. B. Thomas, “Stereovision and navigation in building for mobile robots,” IEEE Transaction on Robotics and Automation, vol. 5, no. 6, pp. 792-803, Dec. 1989.