Research News: A Geometric Approach to Passive Localisation
Theofilos Triommatis, PhD student at the Distributed Algorithms CDT, introduces his latest research paper.
Summary
Nowadays, there is extensive work on computer science and engineering; to make autonomous vehicles with passive sensors safe and efficient in their decision-making. Passive sensors cannot actively scan, but they pick up transmissions and provide the angle of the source's direction with an error as they are not accurate. In this paper, a novel framework is proposed to examine the decision making of m-mobile passive sensors as their objective is to localise the position of k-static emitters inside a given area. The framework replaces the statistical analysis of sensor management by over-approximating geometric objectives. We analyse the emergent behaviour and show the robustness of the proposed algorithms.
Importance of the research
The geometrical approach can be useful at the time when emitters have been detected for the first time. At such a moment, it is hard to make accurate predictions on the emitters' positions because there is not enough gathered data to form a reliable probability distribution or belief. In other words, the bounds on the estimates can be very large which increases the probability of error. This approach provides reliable polygonal bounds on the location of an emitter by only considering the geometry of the problem without using a prior distribution - thereby reducing the reliance on initial assumptions and initial data. We develop an approach based purely on the geometry of the search task; an approach that can complement more sophisticated methods based on Bayesian inference by bounding regions being considered for further processing. In conventional approaches, the uncertainties in the emitter locations are treated as a probabilistic problem. Often, minimising uncertainties involves Markov decision processes to create decision-making mechanisms with various metrics. Another aspect studied is the Emergent behaviour which provides a helpful perspective to study multi-agent systems in various settings. When we have a system with simple rules, we may not be able to predict the outcome of an initial state (unless we simulate it step by step), due to the combinatorial explosion of possible outcomes. This means that understanding the emergent behaviour leads to efficient and reliable algorithms for a large number of initial conditions even in such complex systems.
What comes next
For future work, the centralised approach presented here is closely related to the problem of gossiping as it allows the information of one sensor to become public knowledge. An interesting idea would be to study this problem in temporal graphs as the sensors move and may get out of communication range. In addition, it would be interesting to study the decision-making mechanisms presented in this paper in a distributed rather than a centralised environment.
Please revist this article in mid-July to link through to full paper on IEEE Xplore.
This article belongs to the CDT's Fusion 2022 series. Please review our other Fusion conference paper overviews.