Remote Exploration With Robotic Networks: Queue-Aware Autonomy and Collaborative Localization- [electronic resource]
Remote Exploration With Robotic Networks: Queue-Aware Autonomy and Collaborative Localization- [electronic resource]
- 자료유형
- 학위논문파일 국외
- 최종처리일시
- 20240214101644
- ISBN
- 9798380106740
- DDC
- 629.8
- 저자명
- Clark, Lillian.
- 서명/저자
- Remote Exploration With Robotic Networks: Queue-Aware Autonomy and Collaborative Localization - [electronic resource]
- 발행사항
- [S.l.]: : University of Southern California., 2023
- 발행사항
- Ann Arbor : : ProQuest Dissertations & Theses,, 2023
- 형태사항
- 1 online resource(132 p.)
- 주기사항
- Source: Dissertations Abstracts International, Volume: 85-02, Section: B.
- 주기사항
- Advisor: Krishnamachari, Bhaskar;Psounis, Kostas.
- 학위논문주기
- Thesis (Ph.D.)--University of Southern California, 2023.
- 사용제한주기
- This item must not be sold to any third party vendors.
- 초록/해제
- 요약Lunar and planetary exploration puts stringent requirements on a robotic system, including high reliability, accurate localization and mapping, and the ability to operate and communicate findings with a remote base station despite the lack of existing infrastructure. Robotic networks are well-suited for operation in these harsh, remote environments because the system is robust to the failure of a single robot and agents can communicate and collaborate. In this dissertation, we identify four key problems in the field of networked robotic exploration and provide the necessary solutions to meet the end goal of enabling exploration and mapping in harsh, GPS-denied, communication-restricted environments with a team of mobile robots. We focus on two subdomains in robotic network research: communication and localization.First, we consider that while connectivity is necessary for communicating exploration data, strictly maintaining connectivity can limit exploration. Thus, we propose a queue-aware distributed controller which more flexibly approaches connectivity by focusing on time-average constraints. This improves exploration efficiency without sacrificing timely data transfer.Second, we note that predicting connectivity, which is a key component of queue-aware exploration, is challenging in unknown environments with obstacles that prevent line-of-sight and significantly attenuate signal strength. Thus, we propose a data-driven approach to signal strength prediction which combines the strengths of well-known models of signal propagation phenomena (e.g. shadowing, reflection, diffraction) and machine learning, and can adapt online to new environments. This leads to accurate signal strength prediction which enables communication-aware autonomy for the network of robots. Third, we observe that accurate signal strength models can also enable collaborative localization; given the position of a few robots, we can leverage pairwise inter-robot signal strength measurements to determine the positions of all robots. However, the accuracy of this network localization is again stressed by obstacles which prevent line-of-sight. Thus, we propose a centralized algorithm which first infers and extracts the non-line-of-sight component of attenuation. This leads to accurate collaborative localization which is robust to obstacles in unknown environments.Finally, we acknowledge that a centralized approach to collaborative localization has certain disadvantages, namely communication overhead and synchronization. Thus, we propose a distributed approach to localization for a team of robots with coordinated mobility. Our trilateration-based approach reduces the computational complexity of localization and mapping. This distributed algorithm leads to accurate localization and mapping suitable for resource-constrained robots.We validate these algorithms and models in simulated environments which offer fine control of simulated failures and signal noise. We also validate our proposed methods on robotic systems in real-world environments. We test our low-complexity localization algorithm on a network of four resource-constrained wheeled robots with ultra-wideband positioning devices in an indoor environment. We test our signal strength predictive model and our robust network localization algorithm on a network of three wheeled and three quadruped mobile robots in large-scale subterranean environments. We evaluate our solutions with respect to high-level performance metrics including localization accuracy, mapping accuracy, and exploration efficiency. Further, we analyze implementation-focused metrics including complexity, robustness to noise and failures, and scalability to large networks.Our findings support that (1) queue-aware exploration can improve coverage by 12% compared to the state-of-the-art approach to exploration with intermittent connectivity, (2) data-driven models of the propagation environment can improve signal strength prediction accuracy up to 44% compared to a distance-based model, (3) careful non-line-of-sight inference and matrix manipulation can reduce localization error by 45% compared to the state-of-the-art graph-based learning approach, and (4) trilateration-based localization can reduce complexity by an order of magnitude compared to a well-known simultaneous localization and mapping approach. Together, the four proposed solutions in this dissertation enable a team of mobile robots to efficiently (in terms of time and complexity) explore and map remote environments, e.g. the lunar subsurface, while allowing timely data transfer. Timely data transfer mitigates the risk of losing valuable data due to unexpected failures in harsh environments, and thus we advance the field of remote exploration with robotic networks.
- 일반주제명
- Robotics.
- 일반주제명
- Computer engineering.
- 키워드
- Sensor networks
- 키워드
- Machine learning
- 키워드
- Algorithms
- 기타저자
- University of Southern California Electrical Engineering
- 기본자료저록
- Dissertations Abstracts International. 85-02B.
- 기본자료저록
- Dissertation Abstract International
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.