In July the US Army named 5 companies to build prototype autonomous logistics platforms under Project Sustainment: AM General, American Rheinmetall, Carnegie Robotics, HDT Robotics and Stratom. The program is aimed at automating supply distribution in contested environments and taking soldiers off the resupply runs that expose them most, with field testing alongside operational units expected to begin in early 2027.
It is the right problem to attack. It is also only half of the problem.
Resupply is the most predictable thing a unit does
Ammunition, fuel and water move forward on routes and schedules that are, by necessity, repetitive. That predictability is what makes resupply dangerous: an adversary does not have to find the unit if it can find the route. Taking the crew out of that vehicle removes the part of the mission that cannot be replaced.
Autonomy addresses the delivery. On its own it does not address what gets loaded onto the vehicle in the first place.
Automation inherits the quality of its demand signal
An automated resupply chain runs on 3 answers: what to carry, how much, and when. Those answers come from a demand signal, and in most organizations that signal is still assembled from manual logs, range cards and after-action estimates.
That is a workable input for a human planner who knows which units under-report and which over-order. It is a poor input for an automated system, which will execute the number it is given. Automate the delivery while continuing to estimate the consumption and the forecast has not improved. The guess simply arrives faster, with less friction along the way to catch it.
Where an accurate demand signal comes from
Ammunition consumption is measurable at the point where it happens. A shot counter mounted on the weapon records each round as it is fired, which turns consumption from something reported after the fact into something observed as it occurs.
Senseer sensors do this across more than 60 weapon platforms at roughly 98% counting accuracy. The output is not an estimate of what a unit probably fired. It is a per-weapon record of what it did fire, timestamped and attributable, and it reaches the armory and the planner without anyone having to remember to write it down.
What changes when consumption is measured
Forecasting stops being an argument. Resupply quantities can be derived from observed burn rates per weapon, per unit and per training event, rather than from a standard allowance applied uniformly across all of them. Gaps between what was issued and what was fired surface immediately instead of at the next inventory.
For an autonomous logistics platform, that is the difference between carrying the right load and carrying a confident one.
Automating delivery is a genuine advance, and Project Sustainment is worth watching closely. But the vehicle is only ever as useful as the number it is given. Smarter delivery needs smarter data, and that data has to start at the weapon.