What was demonstrated

Safe Pro’s 15 April 2026 release [1] states that its subsidiary, Safe Pro AI, completed a US Army live-fire exercise using NODE-X — a ruggedised edge-compute node — to process drone-acquired electro-optical imagery in real time and deliver confirmed detections of live scattered mines to the exercise commander. The airborne platform was the US Army’s Short-Range Reconnaissance (SRR) Increment 2 quadrotor, a Blue-UAS-listed asset cleared for organic use at company level. The AI model running on NODE-X is SPOTD (Safe Pro Object Threat Detection), reported as trained on approximately 2.6 million drone images drawn from Ukrainian operational data and 47,000 confirmed positive detections spanning anti-personnel mines, anti-vehicle mines, cluster munition submunitions, and ambush loitering munitions.

The technical distinction matters. Prior minefield screening fieldings rely on tethered networks or remote processing nodes, which are vulnerable to electromagnetic denial and introduce decision latency. NODE-X runs inference on the drone operator’s backpack-mounted GPU without any connectivity to rear nodes, returning georeferenced detections within seconds of acquisition. Further exercises are scheduled through the second quarter of 2026 [1].

Munition natures addressed by SPOTD

Safe Pro has not published an itemised list of all 150-plus recognised ordnance types, but the SPOTD corpus is known from prior public material to include the highest-prevalence scatterable natures on the Ukrainian battlefield. These are the Soviet PFM-1 and PFM-1S butterfly anti-personnel mines (liquid explosive, 37 g of VS-6D or equivalent, pressure actuation), the POM-3 seismic-fuzed anti-personnel mine (with a Medalyon seismic fuze and self-destruct logic), the POM-1 and POM-2 scatterable APERS mines, the PTM-1 and PTM-3 scatterable anti-vehicle mines, and a wide range of cluster-munition submunitions including 9N210/9N235 and DPICM-lineage sub-bombs. The corpus also includes ambush-mode loitering munitions (Lancet and Shahed families) which for EOD cueing purposes present as small high-contrast objects in vegetation.

Analysis of effects — what NODE-X does and does not do

For EOD and route-clearance practitioners the system offers two substantive capabilities. First, it reduces the time between drone over-flight and commander decision from hours (imagery collection, rear processing, dissemination) to seconds. Second, because inference runs on the operator’s compute node rather than in a cloud, the system retains function in denied or degraded EMS environments. This matches the tempo demands of Ukrainian-style high-attrition operations, where Russian tactical drones and GPS-denial make rear-link processing unreliable.

What NODE-X does not do is alter the downstream EOD problem. Detection is not identification, and identification is not render-safe. A SPOTD detection classified as “PFM-1” should be treated as a high-probability cueing point only; EOD protocols under AEP-66 (NATO EOD Principles and Minimum Standards) and JSP 364 (UK EOD doctrine) require visual confirmation and nature-level identification before any render-safe procedure. In cluster-munition submunition detections the problem is compounded by the safe-separation and arming status of individual submunitions, which cannot be assessed remotely.

Personnel and safety considerations

The UK and NATO practitioner interest lies in two domains. First, route-clearance doctrine must adapt to handle the markedly higher detection throughput that edge-AI screening generates. A drone-borne SPOTD pass across a battalion frontage will produce more cueing data than organic EOD assets can act upon in real time; this forces a triage discipline that most NATO EOD forces have not yet codified. Second, training standards under CEN Workshop Agreement 15464 and IMAS 09.30 should evolve to include AI-cued search patterns and to formalise the minimum visual confirmation required before a render-safe procedure is initiated off a SPOTD-class detection [2][3].

Collateral risk is also re-weighted. Drone over-flight of a suspected mined area concentrates detection but can also provoke RC or seismic-fuzed activation. Route-clearance planners should treat SPOTD over-flight height and speed as controlled parameters, not as free variables.

Data gaps

• DATA GAP: Exercise location and unit — Safe Pro has not disclosed the specific installation or US Army unit involved, constraining external assessment of detection performance against specific mine densities and terrain types.
• DATA GAP: False-positive rate — precision/recall metrics against the 150-class model in the exercise scenario have not been published.
• DATA GAP: Detection altitude and sensor — EO, IR, or multi-spectral sensor use, and operational altitude band, not stated.
• DATA GAP: Degradation under jamming — no data on SPOTD performance under GNSS denial or RF jamming typical of a Russian-style EW envelope.
• DATA GAP: Integration with US Army Counter-Mine and Counter-IED enterprise — whether NODE-X feeds into the Army’s existing route-clearance planning tools (ENFIRE, RCPS) is not disclosed.