DASA competition: Autonomous sensor management and sensor counter deception

DASA submissions are welcome from the private sector, academia, individuals (i.e. sole traders) and Public Sector Research Establishments (PSREs). In most cases there are no nationality restrictions, however DASA individual competition documents will detail any necessary restrictions.

The total possible funding available for Phase 1 of this competition is £800,000 (excluding VAT).

This competition is looking to fund 6-12 proposals.

Additional funding for further phases to increase TRL may be available. Any future phase, will be open to applications from all innovators and not just those that submitted Phase 1 successful bids.

Autonomous sensor management is the process of deciding and executing the actions that a group of sensors should take in a given scenario to complete a particular ISR task. In the defence context, such objectives may include target tracking, target classification and developing/maintaining situational awareness. In a world of increasing complexity in terms of the number of sensors and capability, there is a need for algorithms which will process information and disseminate instructions to address ISR queries at human or beyond human-tempo.

Full-spectrum, pervasive ISR is one of the Ministry of Defence’s (MOD) five key capability challenges as outlined in the Science and Technology Strategy. To realise the ambition of tackling ISR challenges at scale and at pace, it is vital to achieve a high degree of autonomy in the processes that underpin the ISR enterprise. This environment will be most-often contested with picture compilation schemes which rely on passive situations, which are of limited utility.

Methods which help detect and counter activities designed to deceive the ISR picture are crucial. This is important, not only because adversaries inherently hide their intent and will project false intent, but also because defence scenarios are nearly unique in this respect. Therefore, drawing across techniques from more benign or cooperative situations has limited utility. Techniques must be developed which have the accommodation for potentially deceptive activity built-in.

Dstl’s principal evaluation and testing route for novel algorithms of this type is via Stone Soup. Stone Soup is an open source software framework for the assessment of tracking and state estimation algorithms. This framework is built in a modular way using an object-oriented approach, enabling construction of sophisticated algorithms from lower-level components. The modularity also enables the same components to be applied to many different domains, targets, sensor modalities, and fusion architectures. Crucially, assessments can be made of different algorithms against representative and consistent simulated, recorded, or live data by using the metrics available in the framework. The Stone Soup source code is available on GitHub and its documentation is available on readthedocs.org.

This competition seeks:

• novel algorithms for autonomous sensor management and sensor counter deception for future ISR systems
• new ideas and contributions to the SFM project via the Stone Soup framework
• to build our community of users and developers by working with a diverse range of external partners.

This competition is interested in supporting innovation at low Technology Readiness Level (TRL). It is expected that Phase 1 projects will deliver outputs at TRL 3.

We require solutions to be delivered in a form which can integrate within a broader set of signal and information fusion technology demonstrators, particularly the algorithms available in the Stone Soup framework. We require open, flexible, and modular architectures to enable integration.

Please read the competition document for a full picture of the scope: the above is only an excerpt.

Sensor management is the process of deciding and executing the actions that a sensor or group of sensors will take in order to meet a particular objective. Objectives could include searching for objects of interest, target detection, identification, recognition, tracking or improved situational awareness among many others, including complex multi-objective problems. For example, an uncrewed aerial vehicle may be required to observe an area to find and track objects of interest. The actions required to search for targets are quite different to those required to track, and so competing objectives will have to be balanced.

The actions a sensor could take encompass tasks such as pointing direction, field of view, beam pattern, sensitivity, power status etc. The tasking of sensor location is also of interest, primarily as a dynamic problem where a sensor can be moved during a scenario (e.g. it is attached to a platform such as an uncrewed vehicle), but also in terms of optimising static sensor placement.

Sensor management algorithms are well explored in academic literature, but difficult to implement due to the practicalities of the problem (i.e. large sets of possible actions to optimise over). This means the autonomous sensor management methods applied to real scenarios are still largely heuristic, with humans anticipating new targets and applying context. This is suboptimal and unsustainable in scenarios of increasing complexity and autonomy.

Initial work has been done to experiment with sensor management approaches using the Stone Soup framework. Simple enactments of sensor management are available on the documentation site.

In the first instance, additional sensor management capabilities are required which are compatible with Stone Soup and this existing setup. The second key element of this request is to demonstrate autonomous sensor management by applying the algorithm(s) implemented in simulation. A scenario should centre on a specific objective (or objectives), with sensor actions selected in order to maximise some reward (or minimise cost).

Possible component developments could include:

• reward functions
• cost functions
• sensor management policies
• optimisation approaches
• “taskable” sensor models

Research areas that might help solve this challenge area may include but are not limited to:

• information theory
• game theory
• reinforcement learning

Military sensing is often complicated by the fact that an adversary’s objective is in direct conflict with one’s own. Adversaries will therefore work hard to obscure their intent, for example by hiding, manipulating their signature or deploying decoys and countermeasures. This complicates inference and means any useful ISR picture compilation must be robust to this activity. Furthermore, behaviour like this is not confined to a single domain or single sensor modality. Techniques found in certain scenarios have analogues in others (e.g. radar jamming/optical dazzle).

The SFM project, under a mandate to improve ISR picture compilation, looks to develop and demonstrate methods to counter sensor deception. Furthermore, it seeks to develop techniques to enable an understanding of the impact deceptive strategies have on a situational awareness picture. Impacts may range from an increase in uncertainty, to attempts to inculcate a false conclusion, or promote a decision conducive to the deceiver. Solutions should employ a defined and recognised set of ISR metrics, an example of which can be found here to quantify the impact of sensor deception, as well as delivering strategies to mitigate such deception. Solutions should be Stone Soup compatible and should allow Dstl to derive and compare metrics which quantify the effect of a range of deceptive activities. Dstl technical partners will work with the supplier to arrive at a common understanding and direct how best to deploy novel algorithms in Stone Soup.

We do not seek to be overly prescriptive as regards solutions. We are, however, looking for methods which can be engineered within future ISR and autonomous systems and will eventually deliver benefit to the intelligence community. Preference will therefore be given to research which shows strong potential for exploitation in this direction. Cross-disciplinary research is encouraged.

Ideas that might help solve this challenge area may include (but are not limited to):

• modelling of intent,
• efficient optimisation,
• game theory,
• Bayesian inference,
• scalable inference on graphs,
• high-dimensional sampling methods.

Competition queries including on process, application, commercial, technical and intellectual property aspects should be sent to the DASA Help Centre at accelerator@dstl.gov.uk, quoting the competition title. If you wish receive future updates on this competition, please email the DASA Help Centre.

If you would like help to find a collaboration partner, contact Hazel Biggs, KTM Security & Defence, or Liqun Yang, KTM Sensors.