The PhD will focus on robust multi-sensor fusion and data acquisition strategies for underwater perception underwater. Underwater sensing is particularly challenging as conditions are highly variable and dynamic. Fixed data extraction and analysis framework tend to be very brittle when used on real environments. Rule based techniques have been used to adapt data acquisition with limited success.
The main thrust of the project will be to develop techniques to robustly acquire data and validate that the data gathered is valid and of sufficient quality to perform specific inspection and manipulation tasks. The techniques should include specific algorithms to identify when the raw data is outside the limits of processing algorithms, adapt the vehicle mission to explore the acquisition parameters to improve raw data and select the best sets of modalities to perform sensor fusion based on the reliability index extracted from the raw data.
Crucially, the system should be able to learn the best strategies over time using reinforcement learning and identify the best sets of modalities to use for specific situation on the fly. We expect the student to work at the intersection of computer vision, machine learning and mission planning to develop innovative solutions for robust and adaptive perception.
The system will be validated in simulation first during the early stages of development and then fully integrated into various vehicles for further validation.
The project will be collaborative in nature and you will interact with our team of researchers and other PhDs in our world leading National Robotarium and Edinburgh Centre for Robotics centres (See https://www.edinburgh-robotics.org/ and https://www.hw.ac.uk/uk/research/the-national-robotarium.htm ). We are currently involved in a number of industry inspired research projects in the area of marine robotics and you will have the opportunity to interact with our industrial partners (internships) and demonstrate your technology on state of the art assets.