Despite the considerable effort directed towards AUV navigation, a self-contained solution remains a challenge. AUVs typically require external navigation support or regular surfacing to obtain GPS fixes to limit the error growth that the inertial navigation experiences. However, these options can be undesirable or unavailable in certain scenarios (e.g. deep-water or under-ice deployments). In response to the currently limited AUV navigation, this project will focus on developing robust sonar-based SLAM techniques for high-powered AUVs operating in surface-constrained environments using on-board sensors and efficient algorithms. SLAM is a powerful technique where a robot builds the map of the surrounding environment and concurrently localises itself within that map. Although SLAM has been proven to be effective for mobile robots operating in structured environments, applying these techniques in the highly unstructured underwater domain is immature, with outstanding challenges to be addressed. The candidate will be required to develop robust and computationally feasible real-time sonar-based SLAM algorithms. Upon the successful demonstration of the effectiveness of the algorithms in numerical simulations, the candidate will have the opportunity to test the algorithm in suitable field experiments (using either the Sparus II AUV or the Autosub2KUI, a high-powered AUV specifically developed for under-ice explorations).
Advances in underwater technology and computing power have yielded new possibilities in the underwater domain. For instance, new underwater navigation techniques have become available , including SLAM [2,3]. However, despite significant progress, key outstanding challenges remain when applied to the underwater domain. Issues include a) feature extraction, data association and loop closure, which are incredibly challenging tasks when operating underwater in highly unstructured environments with limited perception ability, and b) scalability to larger areas, where computationally and memory efficient representations are required. The NOC is currently developing Autosub2KUI, a high-powered AUV specifically designed for operations in ice-covered polar regions. The basic sensor suite includes both ranging and imaging sonars, thus enabling multiple navigation options. This project will focus on developing algorithms for online sonar-based SLAM fusing both ranging and imaging sonars whilst mapping the seafloor and the underside of the ice. The candidate will develop reduced complexity navigation and 3D mapping algorithms that scale the applicability of SLAM to large environments. For the feature detection and data association problems, new solutions will be investigated using machine (deep)-learning techniques. The developed algorithm will be evaluated in numerical simulations and integrated on-board the AUV for real-time field experiments.
The INSPIRE DTP programme provides comprehensive personal and professional development training alongside extensive opportunities for students to expand their multi-disciplinary outlook through interactions with a wide network of academic, research and industrial/policy partners. The student will be registered at the University of Southampton and hosted at the National Oceanography Centre. The project will offer training, and the candidate will acquire skills/experience in the following areas:
- Marine robotics and autonomous underwater vehicles
- Perception and navigation
- Bayesian estimation, machine learning, and optimisation techniques
- Development of high-performance software algorithms for robots
- Collaborations with experts in marine technology and ocean sciences, as well as working in research groups
- Participation in field experiments and research cruises (depending on availability/schedule)
Detailed requirements of a successful candidate:
- MSc degree in computer science, automation, robotics, applied mathematics, or a related discipline in science or engineering
- Strong mathematical skills and analytical skills to work at the intersection of several research domains
- Good programming skills (Python, Matlab, or C++)
- Familiarity with the Robot Operating System (ROS) will constitute an advantage
- Experience in the following areas will be considered favourably: estimation, filtering and machine learning
- Fluent use of English (both written and spoken) and excellent communication and teamwork skills
 Paull, Liam, et al. "AUV navigation and localization: A review." IEEE Journal of Oceanic Engineering 39.1 (2013): 131-149.
 Palomer, Albert, Pere Ridao, and David Ribas. "Multibeam 3D underwater SLAM with probabilistic registration." Sensors 16.4 (2016): 560.
 Ribas, David, et al. "Underwater SLAM in a marina environment." 2007 IEEE/RSJ International Conference on Intelligent Robots and Systems. IEEE, 2007.