Non-myopic approaches to sensing and surveying

Lead Research Organisation: University of Liverpool
Department Name: Electrical Engineering and Electronics


This PhD project tackles the development of high quality but efficient multi-sensor non-myopic sensor management algorithms for controlling sensors and unmanned autonomous systems (UAS) such as unmanned aerial vehicles (UAVs) and unmanned ground vehicles (UGVs). These distributed algorithms are needed in order to exploit the ever-growing capability of autonomous systems for security, monitoring and surveying applications. The algorithms are used to make decisions on things such as where the platforms travel, what direction the sensors are pointed in, and what configuration the sensing will take. Sensor management algorithms typically use Bayesian information theoretic approaches to evaluating the "utility" or "value" of different combinations of sensor and platform actions. Even when such action combinations are evaluated over a single discrete time-step ("myopic" approaches), the problem begins to suffer from combinatorial explosion as the number of sensors is increased or the space of actions enlarges. In order to achieve high quality task choices, non-myopic approaches that consider different combinations of actions over multiple time-steps are required; these approaches are able to trade off short-term gain for higher long-term gain. Non-myopic approaches are able to identify and mitigate for real-world challenges such as obstacles (to gathering information). In general, as the number of timesteps over which the solution is optimised increases, the identified solution will approach a globally optimal solution. There is extensive literature on myopic sensor management techniques and some literature on applied non-myopic approaches. A large amount of prior and relevant work also exists in other areas of computer science and control theory such as the travelling salesman problem and receding horizon control. However, the non-myopic sensor management problem remains under-researched, in large part due to the associated computational challenges. Realised algorithms often only operate on short horizons (e.g. 2/3/4 time-steps) and as such there is still a great deal of potential to improve the effectiveness of such approaches. This project will investigate novel formulations of sensor management problems, as well as the use of Sequential Monte Carlo techniques for identifying high quality solutions. Due to the high levels of computational complexity, a key focus of the project is likely to be understanding the potential for exploiting multi-core computing hardware such as GPUs and/or FPGAs. Another area of interest is how such approaches can generate solutions as part of a human-machine teaming concept, in order that they promote trust and transparency and help end-users understand the inherent trade-offs in the optimisation process.


10 25 50

Studentship Projects

Project Reference Relationship Related To Start End Student Name
EP/S023445/1 31/03/2019 29/09/2027
2599527 Studentship EP/S023445/1 30/09/2021 29/09/2025 George Jones