Perception pipeline for remote tunnel inspection rover. McMahon, C. A. Ph.D. Thesis, May, 2019. Accepted: 2019-11-21T16:49:02Z
Perception pipeline for remote tunnel inspection rover [link]Paper  doi  abstract   bibtex   26 downloads  
Architectural structures require routine structural inspections, and these checks are increasingly performed by remote robotic systems. Inspections often meet all of the traditional "three D's" of robotics - Dull, Dirty, and Dangerous - and may be highly suited to automation. Underground tunnel systems present a particularly important opportunity for robots because these systems may have environmental hazards which make human deployment untenable. Unfortunately, tunnels are also a challenging environment for robots because they feature high self-similarity which makes localization along the tunnel axis difficult, in particular for cases where rough terrain renders odometry information unreliable. Here a platform-agnostic pipeline is presented for tunnel mapping and inspection. This system features tools to automatically segment the various concrete surfaces of the tunnel and individually analyze them for damage. A novel feature-based registration routine is also presented to overcome the localization challenges inherent in tunnels. The pipeline was validated using a LiDAR-based sensor tree and its performance in terms of registration and depth mapping accuracy was extensively tested. The registration methods utilized here do not depend on odometry or the use of registration targets and were shown to outperform contemporary approaches which are standard in industry.
@phdthesis{mcmahon_perception_2019,
	type = {Thesis},
	title = {Perception pipeline for remote tunnel inspection rover},
	url = {https://repositories.lib.utexas.edu/handle/2152/78529},
	abstract = {Architectural structures require routine structural inspections, and these checks are increasingly performed by remote robotic systems. Inspections often meet all of the traditional "three D's" of robotics - Dull, Dirty, and Dangerous - and may be highly suited to automation. Underground tunnel systems present a particularly important opportunity for robots because these systems may have environmental hazards which make human deployment untenable. Unfortunately, tunnels are also a challenging environment for robots because they feature high self-similarity which makes localization along the tunnel axis difficult, in particular for cases where rough terrain renders odometry information unreliable. 
 
Here a platform-agnostic pipeline is presented for tunnel mapping and inspection. This system features tools to automatically segment the various concrete surfaces of the tunnel and individually analyze them for damage. A novel feature-based registration routine is also presented to overcome the localization challenges inherent in tunnels. The pipeline was validated using a LiDAR-based sensor tree and its performance in terms of registration and depth mapping accuracy was extensively tested. The registration methods utilized here do not depend on odometry or the use of registration targets and were shown to outperform contemporary approaches which are standard in industry.},
	language = {en},
	urldate = {2020-05-10},
	author = {McMahon, Conor Alexander},
	month = may,
	year = {2019},
	doi = {http://dx.doi.org/10.26153/tsw/5585},
	doi = {http://dx.doi.org/10.26153/tsw/5585},
	note = {Accepted: 2019-11-21T16:49:02Z},
}

Downloads: 26