Dstl Trials AI-Enabled UAS for EOD Ordnance Detection: British Army Integration

Technical Summary

The Defence Science and Technology Laboratory (Dstl) has completed a multi-week live trial of AI-enabled small uncrewed aerial systems (sUAS) for explosive ordnance detection in support of Explosive Ordnance Disposal and Search (EOD&S) operations. The trial was conducted with 33 Engineer Regiment (Explosive Ordnance Disposal and Search) at their Essex training facility, with dozens of replica surface-laid and buried ordnance items — including simulated anti-personnel mines (APM), anti-vehicle mines (AVM), and abandoned explosive remnants of war (ERW) — positioned across varied terrain types including open ground, vegetation-covered areas, and disturbed soil environments.

The sUAS platforms were equipped with multi-spectral and electro-optical sensor payloads capable of capturing ground-level signature data during low-altitude survey passes. Sensor data was transmitted in near-real-time to Army operators who applied AI-enabled target recognition (AITR) models to detect, classify, and geo-locate identified ordnance items. A critical system attribute demonstrated during the trial was the capacity for rapid in-theatre model retraining — enabling the AI to adapt its detection parameters to new threat signatures and environmental conditions without requiring re-development at source.

Analysis of Effects

The operational impact of a deployable AI/UAS ordnance detection capability is significant across three functional areas. First, it substantially reduces time-in-hazard for EOD operators conducting area clearance: automated aerial survey can cue ground teams directly to ordnance locations rather than requiring full manual lane-search of a suspect area. Second, standoff detection removes search personnel from the explosive hazard during the detection phase, restricting high-risk manual activity to confirmed render-safe procedures (RSP) only. Third, AI retraining capability addresses a longstanding limitation of automated detection systems — their susceptibility to performance degradation against novel or modified threat types. The ability to retrain against new ordnance signatures in near-real-time is operationally significant for EOD task groups operating in threat-adaptive environments.

The trial aligns with the direction set in the UK’s Strategic Defence Review (SDR) for increased integration of autonomous systems, AI, and human-machine teaming in combat support roles. Minister of State for the Armed Forces Luke Pollard described the trial as “exactly the kind of innovation the Strategic Defence Review calls for.” Major Mark Fetters noted the system would allow EOD operators to conduct their mission faster while removing personnel from the explosive hazard during initial area assessment.

Personnel and Safety Considerations

From an EOD safety perspective, the trial represents a meaningful step towards compliance with the ALARP (as low as reasonably practicable) principle for search and detection tasks. Manual route clearance and area search operations constitute the highest-risk phase of most EOD tasks, with operator exposure directly correlated to time spent conducting close-proximity search. A sensor-equipped sUAS conducting initial area survey reduces this exposure to near-zero during detection, with human operators entering the hazard area only after ordnance has been cued and approximate location confirmed.

However, the system’s performance against buried ordnance — particularly minimum-metal content mines and improvised explosive devices (IEDs) with low surface signature — is not quantified in the open-source release. False negative rates (missed detections) and false positive rates (unnecessary ground investigation) at operationally relevant detection thresholds are not disclosed. These parameters are critical for determining whether the system is suitable as a primary survey tool or as a cueing aid to complement conventional EOD&S search techniques.

Data Gaps

DATA GAP: Sensor modality not disclosed (electro-optical, infrared, ground-penetrating radar, or multi-spectral combination). DATA GAP: Detection probability (Pd) and false alarm rate (FAR) metrics for buried ordnance items not published. DATA GAP: Maximum sUAS operating altitude and area coverage rate (hectares per hour) not stated. DATA GAP: AI model training dataset composition — specifically whether training data includes real ordnance signatures or synthetic/replica data only. DATA GAP: Procurement timeline and unit cost estimates not available. DATA GAP: Whether the system has been assessed against low-metal content threats (e.g. wooden-box mines, HPST-type IEDs) or is limited to metallic ordnance with higher surface signature.

AI-assisted technical assessment based on open-source material. Not a formal intelligence product. Source: Dstl / GOV.UK, 2 April 2026.