Flexion, created by a former NVIDIA team, builds a “robot brain”: the plan to replace corporate interns

Swiss startup Flexion Robotics (founded by multiple former Nvidia robotics researchers) has developed a "humanoid robot brain" that allows robots to understand and autonomously execute spoken commands like "go downstairs to pick up a package of snacks, take the elevator up, unbox it, and put the snacks into empty drawers on the shelf." The company was founded in January of this year and has raised approximately $50 million in Series A funding, with investors including Nvidia's venture arm NVentures and DST Global.
(Previous context: Speak and let the robot wash dishes and sweep the floor? Figure launches AI model "Helix" targeting household revolution)
(Background supplement: $2.5 billion valuation! Humanoid robot Digit developer goes public via SPAC)

Key Takeaways

• Swiss startup Flexion Robotics, founded by former Nvidia engineers, has built an "AI brain" for scheduling humanoid robots and raised approximately $50 million in Series A funding.
• The robot first learns individual skills like opening doors, climbing stairs, and carrying boxes in simulation, then a master AI model combines these skills to execute spoken commands, with reinforcement learning at every layer.
• ABI Research estimates the robot foundation model market could reach $150 billion by 2036; analysts直言 the value lies in the AI model, not the robot hardware.

For human interns in the office, this is probably bad news. Swiss startup Flexion Robotics recently showed a video to tech media WIRED: a modified Unitree humanoid robot, after receiving a single spoken command, autonomously walked stairs, took an elevator, unboxed a package of snacks, and organized them into empty drawers on a snack shelf. Throughout the entire process, no one was remotely controlling it.

According to tech media WIRED, Flexion was founded in Zurich this January by several former Nvidia robotics researchers, with CEO Nikita Rudin, who previously served as a robotics research scientist at Nvidia. The company just raised approximately $50 million in Series A funding, with investors including Nvidia's own venture arm NVentures, as well as DST Global, Prosus, and Redalpine. A group of people from Nvidia, taking Nvidia's money, to work on the big thing that Nvidia's boss talks about.

The Brain That Assigns Tasks

Most demonstration videos of humanoid robots on the market actually have a human remotely controlling them (known in the industry as teleoperation), which often fails in unfamiliar environments. Flexion's approach is different: first, in computer simulations, they teach the robot individual skills like opening doors, climbing stairs, and carrying boxes one by one. Then, a "master AI model" decides how to combine these skills to execute a given command.

This master model learns by "watching videos." It digests a large number of videos of humans performing various tasks, matches them to the skills it has practiced in simulation, and then executes them in the real world. For example, to walk to the office mailroom, it autonomously deduces which doors to open and which elevator to take. The same system also controls the motors, allowing the robot to walk, swing its limbs, and maintain balance.

Nikita Rudin says the "secret sauce" of this software is extensive use of reinforcement learning (where a computer learns tasks through repeated trial and error). From the top-level master AI model, to the intermediate simulation, down to the bottom-level motor control, every layer is trained using this method. In other words, Flexion is selling "how to think" from start to finish, not "how to move."

The Value Is in the AI Model

Elon Musk and Nvidia CEO Jensen Huang both believe humanoid robots will significantly transform the economy, as they may eventually replace a substantial portion of human labor. But Flexion's demonstration also makes one thing clear: to make humanoid robots truly capable of work, fundamental breakthroughs in AI are needed first.

George Chowdhury, an analyst at ABI Research tracking the humanoid robot market, put it even more bluntly. He estimates that the "robot foundation model" market alone could be worth $150 billion by 2036. Flexion's software can operate across different brands of humanoid robots. In the current landscape of diverse hardware, this kind of "universal brain" actually holds greater commercial value.

"The humanoid robot itself is not interesting or revolutionary; the AI model behind it is."

Flexion plans to use this new funding to establish a headquarters in the Bay Area, USA. Its business model is to charge an annual software licensing fee based on the number of robots.

FAQ

What is Flexion Robotics?

Flexion Robotics is a robotics startup founded in January 2026 in Zurich, Switzerland, by former Nvidia engineers, specializing in "AI brain" software for humanoid robots. The company has raised approximately $50 million in Series A funding, with investors including Nvidia's venture arm NVentures and DST Global.

How is Flexion's humanoid robot different from other demonstrations?

Most humanoid robot demonstrations rely on human remote control behind the scenes and become unreliable in unfamiliar environments. Flexion instead trains individual skills in computer simulations and uses a master AI model to automatically combine and execute spoken commands, with reinforcement learning at every layer, and the system can operate across different robot brands.