Swiss Startup Flexion Shows a Humanoid Robot Doing Office Tasks on Its Own

June 30, 2026

106

A Swiss startup is making the case that the hard part of humanoid robots is not the body but the brain. Flexion Robotics, based in Zurich and founded in early 2024 by a group of former NVIDIA researchers, has introduced an autonomous platform built to let humanoid machines carry out long, multi step tasks in everyday spaces rather than the fenced off cells of a factory. In its debut demonstration, the company showed a robot handling an ordinary office errand on its own, the kind of mundane chore that turns out to be surprisingly difficult for a machine.

The demo was deliberately unglamorous. Given a single instruction to retrieve a parcel of snacks, the robot found the package, moved through a multi floor building using the stairs on the way and an elevator on the way back, dealt with doors along the route, used a tool to open the box, and placed the items onto a shelf in a snack area. Flexion says it completed the mission autonomously, adjusting when parts of the task changed rather than following a fixed script. None of those steps is impressive in isolation, but stringing them together without a human nudging the robot at each junction is exactly the long horizon problem that has held embodied AI back. Real tasks are chains of small decisions, and a robot that nails ninety percent of them still fails the errand if it stumbles on the rest.

What Flexion is selling is the software that holds that chain together. Its platform, called Reflect, combines mission level reasoning that breaks a goal into steps, perception that reads the environment, the motion and whole body control that move the machine, and a runtime layer that handles communication, logging, and safety checks. The pitch is that this stack is hardware agnostic, a brain that can sit on top of the growing field of humanoid bodies being built by other companies. That positioning matters because the robotics industry has poured enormous effort into legs, hands, and actuators, while the intelligence to use them reliably outside a lab has lagged. Flexion is betting the bottleneck is cognition, not chassis, a thesis that traces back to its founders. Chief executive Nikita Rudin and chief technology officer David Hoeller are former NVIDIA researchers who worked on the company’s Isaac Gym and Isaac Lab robot learning and simulation tools, the kind of background that helps explain both the software first approach and the interest from NVIDIA’s own venture arm.

Investors had already backed that thesis with real money before this week’s demonstration. Late last year the company announced a $50 million Series A from a group that includes DST Global Partners and NVentures, the venture arm of NVIDIA, alongside redalpine, Prosus Ventures, and Moonfire, taking its total funding to roughly $57 million. That round reflected how hot the humanoid category has become, with capital chasing the idea that general purpose robots could eventually take on labor in warehouses, offices, and homes. NVIDIA’s presence on the cap table stands out, since the chipmaker has been pushing hard into what it calls physical AI and supplies much of the compute and simulation tooling the field runs on, the same tooling Flexion’s founders helped build.

The honest caveat is that a single polished demo is not a deployed product. Fetching snacks across an office is a meaningful milestone precisely because it is so ordinary, but a controlled demonstration is a long way from a robot that works shift after shift in an unpredictable workplace, where lighting changes, people get in the way, and objects never sit quite where the model expects. Flexion’s own framing, building intelligence for useful autonomous work outside controlled environments, is an ambition rather than a finished claim. The interesting question is how the system holds up when the parcel is in the wrong place, the elevator is busy, or the task shifts in a way the demo did not rehearse.

Even so, the direction is the point. The conversation around humanoids is shifting from how lifelike the hardware looks to whether the software can actually finish a job without supervision, and Flexion is staking out the brain as the part worth solving first. If the company is right that cognition is the missing piece, a hardware agnostic platform that any humanoid maker can adopt could matter more than any single robot. For now it is an early, well funded bet, and the snack run is a small but concrete sign of where the field is trying to go.