Autonomous resupply: robotic logistics for the tactical last mile

The resupply run is one of the most dangerous tasks a soldier performs in a contested environment. Convoy ambushes account for a disproportionate share of non-combat-initiated casualties in modern conflicts; the predictable routes, fixed schedules, and slow-moving vehicles that resupply logistics demands make convoys lucrative targets. At the same time, driver shortages – both in military establishments and in contracted logistics support – reduce the frequency and volume of resupply sorties that units can sustain. Autonomous resupply is the response: unmanned ground vehicles (UGVs), unmanned aerial systems (UAS), and autonomous maritime platforms, coordinated by AI planning software, performing the resupply mission without placing human drivers in the threat envelope.

This is not a distant technology horizon. UGV logistics platforms are in service with the U.S. Army, the Israeli Defense Forces, and several European NATO members. Cargo UAS programs have reached operational trials across multiple theaters. The software architecture required to plan, dispatch, monitor, and integrate these platforms with existing defense logistics systems is the focus of this article.

The last-mile resupply problem

Defense logistics has achieved substantial visibility at the strategic and operational levels – supply chains from manufacturer to theater depot are largely trackable. The last tactical mile remains the most dangerous and least visible segment. From the brigade supply point forward, resupply vehicles must traverse 5–30 km of contested terrain, often at night, with limited communications, and with predictable route patterns that enemy forces exploit.

The human cost is measurable. Analysis of IED incidents in recent conflicts consistently shows that logistics convoys are targeted at two to three times the rate of maneuver elements per vehicle-kilometer traveled – precisely because their routes are more predictable and their payloads are high-value. Removing human drivers from this segment does not eliminate the threat to the platform, but it eliminates the human casualty component of a successful attack.

The second driver is throughput. A human driver requires crew rest, medical support, and relief rotation. An autonomous platform operates continuously within its maintenance cycle. For sustained high-tempo operations – where resupply demand outpaces available drivers – autonomous platforms extend logistics throughput without expanding the human footprint.

Categories of autonomous resupply platforms

Unmanned ground vehicles (UGVs). The mule-class UGV is the primary platform for ground-level tactical resupply. These platforms – exemplified by the General Dynamics MUTT, Milrem THeMIS, and similar designs – carry 200–1,000 kg of supplies over 50–80 km ranges, following a leader vehicle or navigating semi-autonomously along a pre-planned route. Current platforms operate in three modes: tethered follow-the-leader (the UGV follows a designated human or vehicle using visual or RF tracking), waypoint navigation (pre-programmed GPS route with obstacle avoidance), and teleoperation (remote human control via a video link). True autonomous navigation in complex, GPS-contested terrain remains the frontier.

Unmanned aerial systems (UAS). Cargo UAS serve the aerial delivery role – delivering to positions that are inaccessible by ground vehicle, or where ground access requires traversing heavily contested terrain. Rotary-wing cargo UAS (multi-rotor or helicopter-class) currently deliver 10–150 kg per sortie over 30–150 km. Fixed-wing cargo UAS achieve longer ranges (300–600 km) with lower payload. The key operational advantage of aerial delivery is route flexibility: a UAS can approach a forward position from an unexpected vector, reducing predictability. The constraint is payload – aerial delivery is practical for high-priority, low-weight cargo: medical supplies, communications equipment, specific ammunition types, and batteries.

Autonomous maritime resupply. In littoral and island-chain environments, autonomous surface vessels and semi-submersibles provide bulk resupply capability across water routes that would require either air transport (limited payload) or surface ship (vulnerable, high signature). Autonomous maritime platforms carry several tons of cargo, operate at lower signatures than crewed vessels, and can be pre-positioned at anchor points outside contested coastal zones pending dispatch.

Software requirements for autonomous resupply systems

The software stack for an autonomous resupply system is more complex than the platform hardware suggests. The navigation autonomy is one component; the mission planning, monitoring, and integration layer that makes autonomous resupply operationally useful is the larger engineering challenge.

Threat-aware route planning. Route planning for autonomous resupply must incorporate threat intelligence overlays – known IED belts, enemy observation posts, air defense envelopes, and real-time threat reports from maneuver elements. The planner treats these as cost-weighted exclusion zones rather than hard barriers, allowing the algorithm to trade additional distance for reduced threat exposure. Routes are re-evaluated at configurable intervals and when new threat data arrives from the intelligence feed or from the common operating picture.

GPS-denied navigation. Contested environments involve GPS jamming and spoofing. The navigation stack must degrade gracefully: fusing IMU dead reckoning, LiDAR-based simultaneous localization and mapping (SLAM), visual odometry, and pre-loaded terrain elevation data to maintain a position estimate when GPS is unavailable or unreliable. Position uncertainty is tracked explicitly and surfaced to the operator when it exceeds the operational threshold.

Payload management. The platform must track payload composition, weight, and center-of-gravity throughout the mission. Partial delivery at intermediate waypoints changes vehicle dynamics and remaining range. The payload management module reconciles the physical manifest against the logistics system's expected delivery – discrepancies trigger alerts rather than silent overwrite of the record.

Mission abort triggers and handover protocols. Every autonomous mission must have pre-defined abort conditions: loss of command link beyond timeout, detection of hostile RF signatures, platform fault conditions exceeding defined thresholds, or operator abort command. Abort modes include return-to-base, hold-in-place, and controlled concealment shutdown. Handover protocols define how a stranded platform is located and recovered, and how its payload state is reconciled in the logistics system after a failed mission.

C2 integration: connecting autonomous logistics to the common operating picture

Autonomous resupply platforms are only operationally useful if their position, mission state, and payload status are visible to the commanders and logistics officers who depend on them. This requires integration with the command and control architecture – not as an afterthought, but as a core design requirement.

The integration pattern uses Cursor on Target (CoT) event publishing to a TAK server, making autonomous logistics vehicles visible on ATAK and WinTAK clients alongside maneuver elements and ISR feeds. Each platform publishes its position, speed, heading, mission phase (en route, at waypoint, delivering, returning, aborted), and payload status as a CoT event at a configurable update rate. Platforms in abort or fault state publish distinctive CoT types that trigger automatic alerts on subscriber consoles.

Corvus.Head ingests these feeds and displays autonomous logistics assets on the unified common operating picture – the same COP that shows maneuver elements, ISR coverage, threat overlays, and communications nodes. This gives the operations officer a complete picture of both the tactical situation and the logistics support enabling it, without switching between separate applications. Mission commands – dispatch, abort, reroute, handover to a different operator – are issued through the C2 interface using authenticated, encrypted message channels, with full audit logging of every command and its authorization.

The AI planning layer

Demand forecasting. Autonomous resupply is most effective when dispatched predictively – before units reach critical thresholds – rather than reactively. A demand forecasting model trained on historical consumption rates by unit type, mission type, and operational tempo predicts when each forward position will exhaust each supply category. The model accounts for seasonal and weather factors (fuel consumption is higher in cold weather; vehicle maintenance rates spike after high-tempo maneuver). Forecast outputs feed directly into the mission planning layer, triggering resupply sorties with adequate lead time.

Multi-vehicle route optimization. When multiple autonomous platforms are available, the assignment of sorties to platforms and the sequencing of deliveries is a vehicle routing problem (VRP). A VRP solver – updated in near-real-time as threat data, platform availability, and delivery priorities change – minimizes total fleet travel time while respecting platform range, payload capacity, and threat exposure constraints. The solver outputs an optimized mission plan that the operator reviews and authorizes before dispatch.

Adaptive re-planning. Mid-mission events – a platform fault, a new threat report, a change in delivery priority – require rapid re-planning. The AI layer maintains a live mission model and re-runs the VRP solver when trigger conditions are met, presenting the operator with a recommended re-plan and the delta from the current plan. The operator approves, modifies, or rejects the re-plan within a bounded response window.

Integration with existing defense logistics software

Autonomous resupply platforms do not replace the existing logistics software stack – they must integrate with it. The theater logistics ERP (LOGFAS in NATO contexts, or national equivalents) remains the authoritative system for materiel accounting, fleet management, and fuel tracking. Autonomous resupply missions write delivery events to the ERP through a logistics integration gateway that translates between the platform's telemetry format and the ERP's data model.

The integration is bidirectional: the ERP pushes new requisitions and resupply orders to the autonomous mission planning layer, which schedules sorties accordingly. The mission planning layer writes confirmed deliveries, payload discrepancies, and mission exceptions back to the ERP. Property management systems receive updated accountability records automatically – a delivered item transfers custody from the logistics element to the receiving unit as a logged, timestamped event.

Fuel tracking for autonomous platforms themselves requires special handling: unmanned platforms that operate on electric power or hybrid propulsion have different fuel accounting models than wheeled vehicles. The logistics software must support multiple energy accounting models – liters of diesel, kilowatt-hours of battery, hours of generator runtime – within a unified asset tracking framework.

Human oversight requirements and HITL control points

Current doctrine – and most national regulatory frameworks for autonomous military systems – requires human-in-the-loop (HITL) authorization at defined decision points. Autonomous execution is permitted for routine waypoint navigation, obstacle avoidance within defined parameters, and pre-defined abort conditions. Human authorization is required for: mission dispatch (the operator reviews and approves the mission plan before any platform moves), significant route deviation beyond a defined threshold, encounter with an uncharted obstacle that the onboard system cannot safely classify, proximity to a civilian settlement or protected site, and final delivery confirmation at the receiving unit.

A supervisor console supports simultaneous monitoring of 4–12 autonomous platforms, presenting exception-based alerts rather than requiring the operator to actively monitor each vehicle. The console displays a filtered view: normal operation is a background state; anomalies, abort conditions, and HITL decision prompts surface to the foreground with context data and a time-bounded decision prompt. Operator decisions are logged with identity, timestamp, and the state of the mission model at the time of authorization.

As operational confidence in platform autonomy grows – demonstrated through accumulated mission data, red-team testing, and formal verification – HITL requirements at specific control points can be relaxed through an authorized autonomy expansion process. Lethal proximity operations and civilian area entry retain mandatory human authorization in all current frameworks, with no defined pathway to full autonomy for those categories.