An object recognition and pose estimation library for intelligent industrial automation. Allevato, A. D. Ph.D. Thesis, The University of Texas at Austin, May, 2016.
An object recognition and pose estimation library for intelligent industrial automation [link]Paper  doi  abstract   bibtex   
The nuclear-industrial complex is a field characterized by hazardous environments and stringent worker health regulations. Automation is one of the best ways to improve worker health, but many of the work-intensive tasks in the nuclear industry are difficult to automate using rigid industrial manipulators, which are often treated as glorified assembly lines. This thesis presents the idea of intelligent industrial automation, or IIA, as a way to implement automation in diverse and uncertain environments, and shows that robust computer vision is a key technology in achieving deployable IIA. Furthermore, with recent advances in the field of computer vision, including machine-learning based techniques, the time is better than ever for groups such as the Department of Energy (DOE) to implement computer vision and IIA in their processes. A modular software framework for object recognition and pose estimation (ORP) is developed and incorporated into three laboratory demonstrations, each of which represents a different capability relevant to DOE. By using well-proven computer vision techniques and libraries, ORP enables robust task completion in domains that would have previously been impossible without human supervision or custom mechanical designs (such as task-specific fixtures). A vision-enabled manipulation system is shown to reliably pick and place small weapon detonator components 98% of the time, making it an ideal candidate for machine tending. A remote inspection and inventory system shows the ability to visually detect the position of nuclear material storage canisters with a standard deviation under 1 mm, allowing it to detect cans that have been moved or tampered with. Finally, using vision, an automated glovebox mixed-waste sorting system is able to sort small objects, which begin in a random configuration, into three containers based on their color (a surrogate for radiation signature) with 94.6% accuracy. All three demonstrations proceed autonomously, suggesting that implementing IIA can result in significant improvements in worker safety and productivity at DOE complex sites.
@phdthesis{allevato_object_2016,
	type = {Thesis},
	title = {An object recognition and pose estimation library for intelligent industrial automation},
	url = {https://repositories.lib.utexas.edu/handle/2152/39369},
	abstract = {The nuclear-industrial complex is a field characterized by hazardous environments and stringent worker health regulations. Automation is one of the best ways to improve worker health, but many of the work-intensive tasks in the nuclear industry are difficult to automate using rigid industrial manipulators, which are often treated as glorified assembly lines. This thesis presents the idea of intelligent industrial automation, or IIA, as a way to implement automation in diverse and uncertain environments, and shows that robust computer vision is a key technology in achieving deployable IIA. Furthermore, with recent advances in the field of computer vision, including machine-learning based techniques, the time is better than ever for groups such as the Department of Energy (DOE) to implement computer vision and IIA in their processes. A modular software framework for object recognition and pose estimation (ORP) is developed and incorporated into three laboratory demonstrations, each of which represents a different capability relevant to DOE. By using well-proven computer vision techniques and libraries, ORP enables robust task completion in domains that would have previously been impossible without human supervision or custom mechanical designs (such as task-specific fixtures). A vision-enabled manipulation system is shown to reliably pick and place small weapon detonator components 98\% of the time, making it an ideal candidate for machine tending. A remote inspection and inventory system shows the ability to visually detect the position of nuclear material storage canisters with a standard deviation under 1 mm, allowing it to detect cans that have been moved or tampered with. Finally, using vision, an automated glovebox mixed-waste sorting system is able to sort small objects, which begin in a random configuration, into three containers based on their color (a surrogate for radiation signature) with 94.6\% accuracy. All three demonstrations proceed autonomously, suggesting that implementing IIA can result in significant improvements in worker safety and productivity at DOE complex sites.},
	language = {en},
	urldate = {2017-11-12},
	school = {The University of Texas at Austin},
	author = {Allevato, Adam David},
	month = may,
	year = {2016},
	doi = {10.15781/T2GH9B905},
}

Downloads: 0