News article

Tackling deceptive targets through autonomous systems

25/02/2025
Back to all news

How do you manage autonomous sensor platforms in the presence of deceptive targets? That was the challenge set by the Defence and Security Accelerator (DASA), part of the MOD that supports exploitable innovation for a safer future. Our response? A system called SENTINEL (SENsor management with Target INtent Enhanced Learning).

How do you manage autonomous sensor platforms in the presence of deceptive targets?

That was the challenge set by the Defence and Security Accelerator (DASA), part of the MOD that supports exploitable innovation for a safer future. Using our experience and expertise in autonomous sensor management and sensor counter deception, our teams got to work to find the answer.

Our response was a system called SENTINEL (SENsor management with Target INtent Enhanced Learning).

SENTINEL is a sensor management system using reinforcement learning to develop target tracking policies and predict target intentions, even when their true intent is disguised. For example, a person attempting to mask their actions from intelligence, surveillance and reconnaissance activity.

By comparing predicted routes with the true path of a target, the system can detect deceptive behaviour and account for it. This optimises the sensing platform’s location, giving it a stronger chance to predict the target’s eventual destination.

From pedestrians in urban environments to predicting missile targets, SENTINEL can be used in all sorts of complex situations.

How our new autonomous sensor management works

A map-based tracking example was used to bring SENTINEL to life. We used Open Street Maps, to represent a complex environment and generate target tracks. SENTINEL uses these tracks to build a model that predicts the target’s destination based on its current route. It confines the potential destination to a short list of destinations of interest, then classifies the intended destination using a Long Short-Term Memory (LSTM) model.

For each potential destination, SENTINEL predicts the likely route the target could take, providing a ‘belief state’ that updates as more observations are gathered. By reducing the probability for routes where the target is not detected, SENTINEL can deprioritise these destinations and identify the target’s true track.

Using Multi-Agent Reinforcement Learning (MARL), our sensor management system can be trained to effectively track targets, even when they’re being deceptive. And better still, can be even more deceptive than the target itself by modelling the target’s behaviour.

Learn more about our autonomous software for deceptive tracking.

To find out more about SENTINEL and how it can support deceptive security techniques drop us a line today.