Research News: Non-Myopic Sensor Control for Target Search and Track Using a Sample-Based GOSPA Implementation

Published on

Data Science RS

Dr Marcel Hernandez, Senior Data Scientist with the Signal Processing Group, recently submitted his co-authored paper entitled "Non-Myopic Sensor Control for Target Search and Track Using a Sample-Based GOSPA Implementation" to IEEE Transactions on Aerospace and Electronic Systems


Background

The paper is concerned with managing the operation of sensor resources (e.g., radar, imaging cameras etc.) in order to detect and track evasive targets. The performance metric employed (the "GOSPA" metric) balances the trade-off between inferring the correct number of targets and estimating the states of targets with high accuracy. The optimisation of the sensor actions considers the impact across multiple future time steps (i.e., performs "non-myopic" planning). The paper develops a computationally efficient approach to perform this non-myopic planning in order to minimise the GOSPA performance metric. 

Importance of the research

The GOSPA metric is established as a key performance metric in application concerned with detecting and tracking evasive targets. However, using the metric as a basis for optimising the allocation of sensor resources is computationally challenging. As a result, the metric had previously only been used to perform myopic planning in simply problems with only limited uncertainty (e.g., regarding the locations of potential targets). The approach developed in the paper uses a combination of exact mathematical derivations and sampling of highly uncertain parameters (e.g., the target state) to efficiently calculate the GOSPA metric. This allows the approach to be used in generating non-myopic plans that consider the future impact of current actions. The approach can be used in a broader range of scenarios than previously possible, including coordinated search across a large geographical area. This has allowed the technique to be used to determine scheduling tables for the Tenerife-based Liverpool Telescope (owned by John Moores University) in order to track satellites of interest. 

What comes next?

The research was funded by the Defence Science and Technology Laboratory (Dstl) “Fusion and Information Theory” project. The approach is being extended for ongoing work under the Dstl funded “Reactive Intelligence, Surveillance and Reconnaissance” project that has the challenging objective of scheduling a large number of sensor resources to fulfil a large number of intelligence requests. This extension combines ideas developed for this paper (e.g., non-myopic planning) with approaches developed for game playing algorithms that have advanced to the point of being able to defeat the best human players at complex games such as Chess and Go.

The paper has been accepted for publication in IEEE Transactions on Aerospace and Electronic Systems. A preprint is available at arXiv