Global Logistics Facility Tech Weekly

This week I chose five stories around one question: how much of the decision to let a machine make when it works next to people.

Agility Robotics of the United States unveiled its Digit 5 humanoid on 15 September. It is the company's first model designed to work in the same space as people without safety fencing. AI and multiple sensors detect people nearby, and the robot steers around them, stops or sits down depending on the situation. Lights and sound signal what it will do next, and a safety controller separate from the main control system triggers a response when someone comes too close. It lifts up to about 23kg, 40 per cent more than its predecessor, and reaches about 2.2 metres. Ninety minutes of work needs nine minutes of charging, giving more than 20 hours of operation a day.

What I noted was not the specification but the fact that safety is now settled inside the machine's own control system. The old approach separated people from machines with fencing and constrained the human route by design. Digit 5 moves that line into the machine's judgement. Chief executive Peggy Johnson said the robot was "built to the exact requirements our customers gave us after three years of Digit 4 working on their production floors." Digit 4 has logged more than 65,000 hours at customers including GXO, Schaeffler, Amazon and Toyota's Canadian manufacturing arm.

The supply side has entered mass production. UBTECH of China started up a humanoid plant in Liuzhou, Guangxi, on 12 September with annual capacity of more than 10,000 units. The plant covers 14,000 square metres and is designed to complete one robot every 10 minutes. Several models run down the same line, humanoids themselves handle unloading and material feeding, and collaborative arms drive the more than 2,000 screws each unit requires. The capacity to supply is falling into place.

So where does one start? On 17 September Gartner set out four stages of AI in logistics facilities. The first is more advanced conventional optimisation; the second is generative AI that produces work procedures and exception-handling on the spot. The third is AI agents that recommend or partly execute multi-step work under human supervision, and the fourth is physical AI that actually handles goods. Gartner advises starting with proven applications such as labour forecasting and slotting, and expanding from there.

These four stages correspond to GPIM, the five-stage method GROUND uses to measure the operational maturity of a logistics facility. Gartner's first and second stages map to GPIM Stage 2, Visualisation, and Stage 4, Optimisation; its fourth stage, physical AI, maps to Stage 3, Automation. Only the third stage, AI agents, sits just short of GPIM Stage 5, Autonomy. What matters is the condition Gartner attaches to that third stage: human supervision. The line Digit 5 draws in its safety controller is the same line Gartner asks for in software. People decide how much the machine decides. That is what separates GPIM Stage 4 from Stage 5.

Before the line can be drawn, there has to be somewhere to test. Geekplus of China opened its first European facility of this kind in Düsseldorf on 17 September. More than 150 people attended, including representatives of GXO, DSV and LPP Logistics. Pallet-, tote- and shelf-level transport methods, humanoids and control software can all be compared in one place. Suppliers are beginning to provide somewhere to weigh the options before buying.

Apply this to logistics facilities in Japan. Machines that work 20 hours a day without fencing will be available to buy within a few years. But the more machines there are, the more there is to decide. Machines from different manufacturers, and work done by hand: can one piece of software tell both what order to work in and where to send things? And has someone drawn the line, process by process, on how much the AI decides and where people take over? GROUND's GWES puts that instruction-issuing function and the setting of those limits at the centre of the product, not as something added later. In Japan, where labour shortages are severe and high-variety goods flow through the same building, I believe the precision of that line will show up as a difference in productivity.