The business case for humanoid robots in enterprise settings hinges on their ability to handle diverse, unstructured tasks without constant human guidance. Most current demos rely on teleoperation—a person controlling each movement remotely—which fails when the robot encounters unfamiliar surroundings. Flexion Robotics, a Swiss startup founded by former Nvidia robotics researchers, claims its AI-driven system solves this by training humanoids in simulation and letting a master algorithm compose learned skills autonomously.
How Flexion's AI Works
According to WIRED, Flexion's approach combines multiple AI layers. The main model digests videos of humans performing tasks to understand the required sequence. It then matches those learned skills—acquired through simulation—to the real-world environment. The system also controls the robot's motors for walking, limb movement, and balance. Nikita Rudin, cofounder and CEO of Flexion and a former robotics research scientist at Nvidia, told WIRED that the software's "secret ingredient" is its extensive use of reinforcement learning, which trains computers through trial and error. Every layer, from the master AI model down to motor control, uses this method.
This differs sharply from typical humanoid demos, which rely on teleoperation. Flexion trains its robots in simulation with limited human instruction, making the behavior robust in novel settings. The startup is collaborating with multiple robotics companies and says the software works across different humanoid forms, increasing its commercial value.
The Demo in Action
Flexion demonstrated its system on a modified Unitree humanoid robot. According to WIRED, the robot received the following command: “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 executed the entire sequence autonomously, showing it could open doors, climb stairs, and operate an elevator—skills that are difficult to hard-code.
| Approach | Training Method | Adaptability to New Environments | Human Involvement |
|---|---|---|---|
| Traditional humanoid teleoperation | Human-controlled in real time | Low – fails in unfamiliar settings | High – constant operator required |
| Flexion AI-based autonomy | Simulation + reinforcement learning | High – skills generalize from simulation to real world | Minimal – limited human instruction |
Market Potential and Expert Views
Tech leaders like Elon Musk and Jensen Huang have argued that humanoids will profoundly impact the economy by replacing human labor. WIRED reports that Flexion's demo underscores the need for fundamental AI advances to make that vision real. George Chowdhury, an analyst with ABI Research who covers the humanoid market, told WIRED: “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.
“The humanoid itself isn’t the interesting, revolutionary thing, rather it’s the AI models that back them.” —George Chowdhury, ABI Research
Chowdhury added that Flexion will need to work closely with hardware manufacturers to succeed and faces fierce competition. But without the ability to program humanoids in the way Flexion demonstrates, he said, “there isn’t really a market here.”
Implications for Enterprise Automation
For CTOs and supply chain technology managers, Flexion's approach suggests that humanoid robots could soon perform tasks like package handling, floor delivery, and warehouse sorting without custom programming for each environment. The simulation-first training method reduces the need for expensive real-world trials and enables rapid scaling across different robot platforms. However, competition is intense, and the technology must still prove reliable in production settings. Flexion's partnership strategy with multiple humanoid manufacturers may accelerate adoption, but the market for robot foundation models is nascent. The $150 billion forecast by ABI Research indicates significant long-term potential, but near-term ROI will depend on whether Flexion's AI can consistently handle the unpredictable conditions of real logistics floors.
Sources:
