Flexion Robotics Builds AI-Powered Humanoid

What did Flexion Robotics just demonstrate?

Flexion Robotics is a Swiss startup founded by former Nvidia robotics researchers that trains humanoid robots to perform complex, multi-step tasks using simulation and reinforcement learning. The company published a demo video showing a modified Unitree humanoid robot operating entirely on its own after receiving one natural-language command.

That command was: "A parcel with snacks has been delivered for Flexion. Retrieve it using the stairs and come up using the elevator. Then unpack it and place the items into the empty drawer on the shelf in the snack area." The robot completed every step without a human operator, as Wired reported via dnyuz.com.

How does Flexion's training system work?

Most humanoid robot demos rely on teleoperation — a human controls the robot's movements from behind the scenes. That approach breaks down when the robot encounters unfamiliar environments. Flexion's system is built differently.

The main AI model learns by watching videos of humans performing tasks. It then matches those observations to skills the robot has already practiced in simulation. To reach a mailroom, for example, the model figures out it needs to open doors and call an elevator. A separate layer of the software controls the robot's motors — handling walking, limb movement, and balance.

The system combines these layers through a master AI algorithm that decides which learned skills to apply in sequence.

You might also like

AI Boom Drives Global Memory Chip Shortage 2026

Taiwan Raids Super Micro in Nvidia Chip

Amazon Eyes Alt AI Models After Anthropic Price

Kunlunxin Eyes $50B Hong Kong IPO Valuation

What is the "secret ingredient" Flexion's CEO describes?

According to Nikita Rudin, Flexion's cofounder and CEO and a former robotics research scientist at Nvidia, the system's "secret ingredient" is its extensive use of reinforcement learning — a training method where computers master tasks through trial and error. Every layer of the software uses this approach, from the master AI model down to motor control.

Here's what we know so far: this is a meaningful departure from how most robotics teams train their systems, which typically require far more direct human instruction.

Why does the AI software matter more than the robot hardware?

George Chowdhury, an analyst at ABI Research who follows the humanoid market, put it plainly: "The humanoid itself isn't the interesting, revolutionary thing, rather it's the AI models that back them."

ABI Research estimates the market for robot foundation models could be worth $150 billion by 2036. That figure frames why software companies like Flexion are drawing attention — the hardware is becoming a commodity, but the AI that runs it is not. This dynamic mirrors broader trends in AI model access and the race to control the software layer.

Who are Flexion's competitors and partners?

Rudin says Flexion is already collaborating with several robotics companies. He also notes the software works across different humanoid form factors — a significant commercial advantage given how many competing robot platforms exist on the market today.

Chowdhury acknowledged the competitive pressure Flexion faces. He said the company will need to work closely with hardware manufacturers to succeed. But he also made the stakes clear: without the ability to program humanoids the way Flexion demonstrates, "there isn't really a market here."

The pressure to build capable AI for physical robots is intensifying across the industry, similar to how AI expansion is reshaping hiring and investment decisions at major tech firms.

Key facts about Flexion Robotics at a glance

What does this mean for humanoid robots replacing workers?

Tech leaders including Elon Musk and Jensen Huang have argued that humanoids could eventually replace a significant portion of human labor. Flexion's demo reflects how much AI development still needs to happen before that becomes possible.

The robot in the demo completed a task that any office intern could handle — but doing so required a sophisticated stack of AI systems trained across simulation, video observation, and motor control. The gap between a compelling demo and reliable deployment at scale remains a core challenge for the entire humanoid robot field.

Flexion's software-first approach — building AI that works across multiple robot platforms — positions it as an infrastructure play in a market that ABI Research values at $150 billion by 2036, according to technewstube.com's coverage.

The most confirmed next step from the sources: Flexion is actively collaborating with multiple robotics companies and says its software is already compatible with different humanoid form factors.