融合天花板與地面互補式資訊地圖於複雜環境之定位導航系統

Complex Environment Localization System Using Complementary Ceiling and Ground Map Information

余奇安(Chee-An Yu)

指導教授 ： 傅立成

國立臺灣大學/電機資訊學院/電機工程學研究所/碩士(2020年)

https://doi.org/10.6342/NTU202002645

摘要

近年來，由於少子化、人口老化與勞動力短缺等問題，透過機器人提供老人照護，居家陪伴或是補充勞動力的研究逐漸顯示其重要性。在高齡化社會中，機器人為以上問題提供多元的解決方案。為了使機器人有完善的移動社交能力，機器人需要有強健的感知能力，定位能力與導航能力。其中定位能力又是機器人在移動導航過程中最核心與基礎的功能。機器人要能夠理解自己與環境座標的相對位置（localization），在未知環境中要有建地圖的能力(mapping)，或同步定位與建地圖（simultaneous localization and mapping, SLAM）同時估計自己在環境中的座標位置與建立環境地圖。在實際應用的場景中，機器人往往是面對動態環境甚至人來人往的複雜多雜訊環境中。因此需要一個有效能使機器人免於這些雜訊影響，達到精準定位的系統。此外，在多人擁擠的環境中，機器人也有可能因為不可預期的人為干擾與移動而迷失原本的定位，因此有關機器人定位失準與迷失的檢測系統與重新定位的演算法機制也是機器人真正能在人群中移動導航與人互動不可或缺的因素。
本篇研究中，我們提出了一個基於天花板3D點雲感測資料與地面雷射感知互補式資訊系統來達到機器人在高複雜度的動態室內環境與擁擠人群中的強健定位能力。透過天花板幾何特徵與相比於地面不易受動態環境影響的特性，取出較穩定且不易隨時間而改變的環境特徵達到強健的定位功能。同時利用來自地面的障礙物與對人比較直觀的地圖達成避障與移動導航。透過天花板地圖與地面地圖的整合架構，不僅定位能獲得改善，移動導航的能力也進而獲得提升。得利於此系統，即使在地面高複雜與動態不確定性的情形下，甚至在移動過程遭受外力的干擾使機器人的真實位置與機器人認知的位置有所不同的問題發生時（綁架問題），機器人仍可達成強健而穩定的室內定位功能。

關鍵字

3D 點雲 ； 複雜環境定位 ； 感測器融合 ； 天花板定位與建圖

並列摘要

In recent years, due to declining of birthrate, aging of the population, and the shortage of labor power, the research on robotics gradually shows its importance and provides various solutions. For example, robots have become part of the family to take care of the elderly or accompany them to enrich their life. On the other hand, robots like autonomous ground vehicles also use their mobility to accelerate the product process of the automation system. In order to bring robots into our daily life, the mobility of the robot is the core and vital ability. For the new and uncertain environment, the robot should have the ability of storing sensing information from environments as a map for future usage. In real application fields, robots usually face crowd people in dynamic and uncertain environments, which are challenging to model and sometimes result in localization failure. Besides, during the interaction process, the robot may suffer unexpected movement by others or dramatically changing of the environments and thus loses track of the robot itself. Therefore, a system equipped with losing tracking detection and recovery behavior mechanism for localization is the essential ability for the mobile robot that can function soundly in dynamic and complex environments.
In this thesis, we propose a robust localization system using complementary information from the ceiling and ground environment, which can be applied to dynamic and complex environments. The ceiling perception provides a robust and time-invariant feature for robot localization, and the ground perception allows the robot to navigate and avoid obstacles. Through the fusion of the two types of complementary information, not only the localization ability from ceiling perception and ground map can be enhanced, but also the navigation performance can be improved significantly. Furthermore, without articulated landmarks by humans, the proposed system can map the ceiling and ground environments efficiently and reuse information effectively for robot localization. Besides, through the design of integrating of both the ceiling information and ground map, the kidnap problem can also be detected effectively, and the pose can be recovered more efficiently. With the help of the proposed system, the robot can localize itself in the dynamic and cluttered environment with the ability to recover from the kidnap problem.

並列關鍵字

Point Cloud Feature ； Complex Environment Localization ； Ceiling Localization and Mapping ； Multi-Sensor Fusion

參考文獻

[1] S.Thrun, W.Burgard, and D.Fox, Probabilistic Robotics. 1999.

[2] T. Collins, J. J. Collins, and C. Ryan, “Occupancy grid mapping: An empirical evaluation,” in Mediterranean Conference on Control and Automation, 2007.

[3] T. Collins, J. J. Collins, M. Mansfield, and S. O’Sullivan, “Evaluating techniques for resolving redundant information and specularity in occupancy grids,” in Lecture Notes in Computer Science, vol. 3809, pp. 235–244, 2005.

[4] T. Le and Y. Duan, “PointGrid: A Deep Network for 3D Shape Understand ing,” in IEEE Computer Society Conference on Computer Vision and Pattern Recognition, pp. 9204–9214, 2018.

[5] R. Valencia, J. Saarinen, H. And reasson, J. Vallve, J. And rade-Cetto, and A. J. Lilienthal, “Localization in highly dynamic environments using dual-timescale NDT-MCL,” in IEEE International Conference on Robotics and Automation, pp. 3956–3962, 2014.