IN Brief:
- GXO has moved an AI-driven autonomous industrial truck into live operation at a high-volume warehouse in Épinoy, France.
- The pilot combines KION vehicle systems with digital-twin modelling, AI perception, and warehouse orchestration developed alongside NVIDIA and Accenture.
- The next test is whether physical AI can be rolled out repeatably across existing brownfield warehouse operations at commercial scale.
GXO has started operating an AI-driven autonomous industrial truck at its facility in Épinoy, France, taking physical AI out of simulation and into a live warehouse environment. The pilot is being run with KION and sits within a wider collaboration involving NVIDIA and Accenture, with the aim of proving whether next-generation autonomous material handling can deliver measurable gains in productivity, scalability, safety, and operating cost.
The significance of the project lies in where it is running. Épinoy is not a test hall built around a clean automation concept, but an active, high-volume site where the autonomous vehicle has to work inside an existing operating model. Before deployment, the warehouse was captured with spatial scanners and converted into a digital twin. From there, the truck was trained to complete full transport missions autonomously, using ceiling-mounted and onboard AI-enabled cameras to identify pallets and move them to defined drop locations.
KION says the vehicle is running alongside personnel and manually operated forklifts, which matters more than the headline novelty of autonomy. Many warehouse operators no longer need proof that an autonomous truck can move in isolation. What they need is evidence that it can work in mixed traffic, inside inherited layouts, and without demanding a full rebuild of site processes. GXO’s role here is effectively to serve as a production testbed for that transition.
The pressure point now shifts from demonstration to repeatability. If the pilot delivers consistent performance in a warehouse that already uses more than 200 manual trucks, it becomes easier to argue that physical AI can be added to existing logistics estates rather than reserved for greenfield automation projects. That is where the commercial case will be decided: not in a conference hall, but in whether these systems can be deployed quickly, safely, and with enough operational discipline to scale across multiple customer contracts.
