A robot in a Martur Fompak automotive seat factory spent January and February 2026 receiving task assignments from an AI agent — not a human operator, not a custom control system, but a software agent running in the cloud. Martur Fompak, which employs more than 7,100 people across 23 production plants in six countries, declined to comment for this article. The AI agent was somewhere else entirely. The connection between them: the public internet.
The proof-of-concept, completed by Humanoid, a UK-based robotics company founded in 2024, is being described by the company as a step toward real deployment. The Robot Report covered the announcement. The Robot Report's primary source is a Humanoid press release. Nothing in that announcement has been independently verified by anyone outside the three companies in the room.
Humanoid — founded by Artem Sokolov, who has self-funded the company with roughly $30 million of his own capital, according to Sifted — has positioned this as its eighth proof-of-concept with a Fortune 500 partner, with Schaeffler, Siemens, Ford, and Martur Fompak among the named partners. Martur Fompak, an automotive seat assembly supplier with stated ambitions to deploy intelligent tote handling across 30 global production facilities — a deployment target, not its current plant count. The HMND 01 Alpha robot used — a torso with arms on a wheeled platform, 300 kilograms, 220 centimeters tall, built for warehouse and logistics work — was linked to SAP Extended Warehouse Management via SAP's Joule agent layer, a middleware product designed to let enterprise systems assign tasks to physical machines. The Register confirmed the architecture: tasks arrived from SAP's system, not from a custom local control system sitting next to the robot.
The robot handled three different tote types within an 8kg dual-arm payload limit. The KinetIQ framework — Humanoid's four-layer AI architecture, with a visual-language model reasoning at 5 to 10 predictions per second and a lower-level controller executing at 30 to 50 hertz — translated task directives into continuous motor commands. Per Humanoid's own account, this was the first time its robot had been controlled by an external enterprise system in a live production environment.
"This proof of concept shows what matters: Humanoid robots operating inside real production environments, connected to enterprise systems and measured against operational standards," Sokolov said in the announcement. "That is the bridge between experimentation and deployment."
The "bridge" framing is nice. What the announcement actually demonstrates is narrower: a robot received task assignments from SAP's warehouse management system over the internet. No local integration. No custom control layer. Whether that matters depends on what you're trying to sell.
Here is the part worth sitting with: the task assignments came from SAP's system. The same infrastructure that routes purchase orders and tracks inventory was routing tasks to a machine on a factory floor. That is, in a specific and literal sense, what enterprise software looks like when it reaches into the physical world. The robotics press will write about the robot. The enterprise software press might write about the middleware. The interesting question is who is actually buying either one — and why.
The cost case against humanoid robots in logistics is not subtle. The International Federation of Robotics published a position paper this month arguing that the high cost of materials, components, and programming complexity renders humanoid robots unaffordable for cost-effective operations at scale. That is the IFR's argument, not a fact — but it is the argument that every warehouse operator considering a seven-figure investment in untested hardware has to contend with. The Goldman Sachs price-drop narrative ($250,000 down to $150,000 per unit, per an interview with Sokolov) is real, but the question is whether the math works at a customer's actual facility with their actual workflow, not on a spreadsheet.
The counterpoint exists and is operational. Agility Robotics' Digit robot has moved over 100,000 totes at a GXO Logistics facility in Georgia, per Agility's own announcement. That is not a proof-of-concept. That is a robot doing work in a building where people also work. It does not mean the economics are solved. It means at least one company in this space has moved beyond the demo-and-announce phase.
Humanoid has not. Eight proofs-of-concept with Fortune 500 partners is a real list. It is also a list of eight companies that have not signed a commercial contract, disclosed a price, or published operational data from a live deployment. The Siemens proof-of-concept — 90-plus percent pick-and-place success rate, 60 tote moves per hour, more than eight hours of continuous autonomous operation — is the most detailed performance data Humanoid has published. It is also a proof-of-concept, not a commercial deployment.
The honest framing for this announcement: a self-funded UK startup with 200-plus engineers ran a two-month pilot at a customer facility, connected its robot to existing enterprise software, and published the results. That is not nothing. It is also not evidence that humanoid robots are operating at commercial scale in automotive logistics. The gap between those two statements is the entire debate about this industry right now.
Humanoid says it is targeting first commercial delivery in early 2027. That timeline — 18 months out — is where the story will actually be decided. The KinetIQ middleware connecting SAP to a robot is genuinely novel. The commercial case for why a warehouse operator should pay for it has not been made yet. Those are two different things, and conflating them is how you end up writing a press release instead of a story.
What happens in that 18 months — whether one of those eight POC partners converts, whether the 100,000-tote milestone at GXO prompts others to move faster, whether the IFR cost critique or the Goldman Sachs price-drop narrative wins the argument inside a procurement meeting — is the actual story. This announcement is the setup.
