Technical Summary

The Defence Science and Technology Laboratory (Dstl) conducted a multi-week trial of Artificial Intelligence (AI)-enabled small Uncrewed Aerial Systems (sUAS) for the detection and classification of landmines and explosive ordnance. The trial was hosted by 33 Engineer Regiment (Explosive Ordnance Disposal and Search) at their base in Wimbish, Essex, and involved dozens of replica mines and ordnance items placed across varied terrain types.

Data captured by onboard sensors was relayed to British Army operators who used AI tools to locate and identify the munitions. The trial demonstrated two core capabilities: automated detection and classification of known ordnance types, and — critically — rapid retraining of AI models to recognise emerging threat types and adapt to different environmental conditions. This retraining capability is the operationally significant finding. Static AI models trained against a fixed threat library become obsolete as adversaries modify or improvise explosive ordnance. The ability to update detection models in the field, against novel threat signatures, is what separates a demonstrator from a deployable capability.

The UK Government has committed to doubling autonomous systems investment from £2 billion to £4 billion during this Parliament. Further trials are planned in 2026 to mature the technology toward procurement of a deployable system that can be issued directly to combat engineers.

Analysis of Effects

Current counter-Explosive Ordnance (EO) detection in the British Army relies on a combination of manual search techniques, metal detectors (Vallon VMH3CS and equivalents), Ground Penetrating Radar (GPR), and vehicle-mounted mine detection systems. These methods are effective but slow, manpower-intensive, and expose search personnel to the threat they are seeking to identify. The standoff provided by sUAS-mounted sensors fundamentally changes the risk calculus: the operator remains at distance while the sensor moves through the threat area.

The AI component addresses the second bottleneck — interpretation. Raw sensor data from electro-optical, infrared, or multispectral sensors generates volumes of imagery that exceed human processing capacity during time-constrained operations. AI classification reduces the sensor-to-decision timeline from hours to minutes, enabling combat engineers to prioritise confirmed or suspected ordnance locations for manual verification and Render Safe Procedures (RSP).

The retraining capability is the distinguishing feature. Ukraine’s conflict has demonstrated that the mine and Improvised Explosive Device (IED) threat evolves continuously — modified anti-tank mines with anti-handling devices, 3D-printed mine casings with minimal metal content, and commercially sourced components that do not match legacy threat signatures. An AI model trained in 2025 against a fixed library of NATO and Warsaw Pact mine types will not reliably detect a modified TM-62 with a non-standard fuze or a locally fabricated pressure plate. Rapid retraining against captured examples or intelligence reports of emerging threats is operationally essential.

Personnel and Safety Considerations

33 Engineer Regiment (EOD&S) is the British Army’s specialist unit for explosive ordnance disposal and military search. The Regiment provides EOD and search support across the full spectrum of operations, from humanitarian demining to high-threat IEDD in contested environments. Their involvement in this trial ensures the technology is being assessed by operators who understand the tactical constraints of search operations — not just the laboratory conditions in which detection algorithms are typically developed.

A sUAS-based detection capability does not replace the requirement for qualified Ammunition Technicians (ATs) or EOD operators to conduct RSP. The technology provides an enhanced search tool that improves area coverage rates and reduces personnel exposure during the detection phase. The render-safe phase — approaching, identifying, and neutralising confirmed ordnance — remains a manual task requiring human judgement about fuze state, energetic condition, and environmental factors that AI cannot currently assess.

Data Gaps

DATA GAP: Sensor modalities — The specific sensor types on the sUAS have not been disclosed. Electro-optical and infrared are probable, but the capability to detect buried ordnance (as opposed to surface-laid mines) would require Ground Penetrating Radar (GPR) or magnetometry, which are not typically carried on small UAS platforms due to weight and power constraints.

DATA GAP: Detection performance — No probability of detection (Pd), false alarm rate (FAR), or area coverage rate metrics have been published. Without these, the operational utility of the system cannot be compared against existing manual search methods or vehicle-mounted systems.

DATA GAP: Platform endurance — sUAS endurance (flight time, range) and the number of platforms required to achieve useful area coverage rates have not been specified. Battery-powered sUAS typically offer 20–40 minutes of flight time, which constrains the area searchable per sortie.

DATA GAP: Industry partners — The commercial or defence industry partners providing the sUAS platform, sensors, and AI software have not been named. This is typical for Dstl trials during the assessment phase but limits the ability to assess technology maturity.