Multi-robot navigation in unknown cluttered environments: application to radiation detection
Date
2021
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
This dissertation pushes the boundaries of autonomous robotic navigation by co-designing the trajectory planning algorithms with a decision-making component. Specifically, it presents navigation strategies for multiple dynamic sensor platforms so that they can safely move in an unknown environment and collectively make a confident decision about the dynamic process they are measuring. The key application area that has been discussed throughout this dissertation is the detection of weak radiological materials using radiation counters of the Gieger-Muller (GM) type. Although quadrotors have been selected as the preferred choice for the sensor platform, most of the algorithms developed are agnostic to the platform choice and can be used with any combination of different robots. ☐ This sensing and detection capability could be the part of national defense agency’s strategy to detect, deter and defeat radiological threats and could facilitate the development of superior technologies to check, monitor or verify the adherence of member states to the nuclear non-proliferation treaty. Not limited to a particular application, such a capability could be used in any specialized search operation in which the dynamic process is a time in-homogeneous counting process, such as detection of the black box of a lost aircraft. ☐ This work leverages the previous worked performed on developing the radiation detection theory and optimal motion strategies for point robots moving in a known/obstacle-free environments and extends it to include full dynamics of the moving platform. These platforms can navigate in a completely unknown environment using real-time onboard perception capabilities. A computationally efficient receding horizon planning strategy generates dynamically feasible motion plans reactively, to steer the robot towards any static or moving destination. Although mission safety takes precedence in such scenarios, the motion of the platform is designed to give best possible decision-making performance. The receding horizon planning approach has been further extended and integrated with navigation function based global planners to provide exactness and convergence guarantees in scenarios where the robot can get trapped in local-minima without the knowledge of the environment. ☐ The complete planning and decision-making pipeline has been extensively tested in simulations as well as indoor and outdoor field experiments using in-house developed quadrotors and 8.2 μCi gamma-ray sources and have been shown to give an order of magnitude improvement against the existing methodologies using similar configuration of radiation sensors while detecting source of comparable intensity. ☐ The dissertation further investigates the effect of relative motion of the source and the sensor platform on to the decision-making accuracy and devices strategies for optimal sensor coordination and formation design (while chasing a target) for improved detection performance. ☐ One limitation of this work is that it relies on the knowledge of the target that carries the source and the intensity of the source itself. Although most autonomous robotic deployments, including this one, is far from being truly and completely autonomous, such a requirement is restrictive and hinders the practical deployment of the system. A possible extension of the work could be to localize the source and calculate the intensity of it using radiation sensors. Discussed further in the last chapter of this dissertation, full 360° directional radiation sensors could possibly be integrated to the developed system for such an extension. This would clear the way for the real-world deployment of the presented system/algorithms.
Description
Keywords
Control theory, Motion planning, Multi-agent coordination, Quadrotor navigation