IN Brief:
- Indra led FaRADAI use cases applying frugal, robust, and explainable artificial intelligence to military operational problems.
- Demonstrated functions include scene interpretation, object identification, relationship analysis, threat assessment, and dynamic generation of safer routes.
- The European Defence Fund programme emphasises constrained computing resources, sensor resilience, human oversight, and sovereign European AI capability.
Indra Group has demonstrated artificial-intelligence functions designed to interpret complex military ground scenes, identify relevant objects, assess potential threats, and generate safer routes when unexpected conditions alter an operational plan.
The work forms part of FaRADAI — Frugal and Robust AI for Defence Advanced Intelligence — a European Defence Fund research programme focused on artificial intelligence able to operate with limited computing resources while remaining robust and sufficiently explainable for use in defence applications.
Indra led several critical use cases within the project. One pilot used ground-based imagery to identify vehicles, personnel, and infrastructure before analysing relationships between those detected objects to support threat assessment. The same architecture could then recalculate safer routes when new circumstances affected the original plan.
The programme also examined sensor resilience under misleading conditions. That is a more difficult problem than demonstrating object recognition against clean imagery because operational sensors may be degraded by weather, obscuration, movement, incomplete coverage, electronic interference, or deliberate attempts to deceive the system.
An AI model can still produce a confident answer when the information reaching it is poor. Robustness therefore involves understanding how the system behaves when its inputs are uncertain or misleading, rather than measuring performance only against carefully selected test data.
FaRADAI’s emphasis on frugal computing addresses another practical constraint. Military AI cannot assume continuous access to large cloud-computing resources. Vehicles, dismounted units, forward headquarters, and isolated sensors may have limited electrical power, constrained processors, intermittent communications, and little bandwidth available to move large quantities of data elsewhere for analysis.
Software able to run closer to the sensor can reduce those dependencies, but the models have to be efficient enough to operate on the hardware available. That shifts part of the engineering challenge from developing the largest possible model towards achieving useful performance within realistic power, memory, thermal, and communications limits.
Indra has also built a human-in-the-loop approach into the demonstrated functions. Artificial intelligence is intended to support human decision-making rather than remove operators from critical choices, particularly when the software is identifying a potential threat or recommending a change of route.
Explainability is central to making that relationship useful. Operators do not necessarily require access to every internal calculation, but they need enough context to judge why a system has highlighted an object, changed its assessment, or proposed another route. Confidence, contributing sensor information, and consistent behaviour all influence whether a machine-generated recommendation deserves operational trust.
The FaRADAI consortium comprises 38 partners from 13 European countries and is coordinated by Greece’s Centre for Research and Technology Hellas. Representatives from European defence ministries and the European Commission reviewed the programme’s results in Thessaloniki, connecting the research with potential government users rather than leaving it entirely within laboratory development.
Indra is also tying the work to technology already used in military situational awareness. Its Local Situational Awareness System and See-Through 360-degree vehicle-vision technology use distributed sensors to improve crew visibility, while the company’s IndraMind platform is intended to provide a sovereign and cyber-resilient intelligence layer able to interpret operational data.
That provides a potential route from research functions into deployed equipment, although a European Defence Fund demonstration is not a production qualification. Integration into an operational platform would still require work around processing hardware, cyber security, sensor interfaces, data governance, environmental resilience, software assurance, and the procedures determining how automated recommendations are used.
Sovereignty is another component of the programme. Defence organisations adopting AI have to consider where models are developed, how sensitive training and operational data are controlled, whether software can be maintained independently, and how changes are tested after a system enters service.
Those issues become more significant as artificial intelligence moves closer to command and sensing functions. A commercially capable model may be unsuitable if a defence customer cannot understand its dependencies, control its data, or maintain it throughout the service life of the platform hosting it.
The route-planning use case provides a useful example of how AI can be introduced without assigning it final authority. Software can process changing sensor information and calculate alternative routes more quickly than a human operator, while the commander retains responsibility for weighing terrain, mission priorities, intelligence quality, rules, and risks that may not be completely represented by the model.
FaRADAI has shown that those technical functions can be combined under controlled programme conditions. The more demanding step is proving that the same behaviour remains reliable when sensors are degraded, computing resources are limited, and operators have little time to determine whether the machine’s recommendation should influence the next decision.
