Introducing Flexion Robotics and Its Innovative Training Method
Humanoid robots may already be able to walk, climb, and occasionally dance, but to become truly human, they are going to need to learn how to perform all sorts of menial chores at work. Flexion Robotics, a Swiss startup founded by ex-Nvidia robotics researchers, believes it has the solution. The company has developed a way to train humanoid robots to execute complex tasks involving simple skills like opening doors, climbing stairs, and carrying boxes. The key is to teach the robots individual skills in simulation, then have a master AI algorithm determine how to combine them.
Most demo videos show humanoids trained for a specific task—such as folding shirts or loading shelves—typically via teleoperation, where a person behind the scenes controls the robot’s movements. However, this approach fails reliably when the robot encounters unfamiliar settings. Flexion claims its system is different and more efficient because it trains robots in simulation with limited human instruction.
How Flexion’s AI System Works
Combining Simulation and Reinforcement Learning
Flexion’s approach works by combining different AI systems. The main AI model learns chores by digesting videos of humans performing various tasks. The software then matches learned skills—acquired in simulation—to those videos and executes them in the real world. For example, to reach the mail room in an office, the model may have learned it needs to open certain doors and use the elevator. The system also controls the robot’s motors, enabling walking, limb movement, and balance.
According to Nikita Rudin, Flexion’s cofounder and CEO—formerly a robotics research scientist at Nvidia—the software’s “secret ingredient” is its extensive use of reinforcement learning, which trains computers to master tasks through trial and error. Each layer of the software, from the master AI model to simulation to motor control, uses this approach.
Real-World Demonstration
The video below shows the software in action: A modified Unitree humanoid robot operates autonomously after receiving the 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.”
Market Potential and Challenges
Economic Implications and Competition
Tech industry leaders like Elon Musk and Jensen Huang argue that humanoids will have a huge economic impact by potentially replacing a significant portion of human labor. However, Flexion’s demonstration highlights that empowering humanoids requires fundamental advances in AI. “The humanoid itself isn’t the interesting, revolutionary thing; rather it’s the AI models that back them,” says George Chowdhury, an analyst with ABI Research, who follows the humanoid market. ABI Research estimates the market for robot foundation models could be worth $150 billion by 2036.
Rudin says Flexion is collaborating with multiple robotics companies and notes that its system works across different humanoid forms. Given the number of systems on the market, this could make the software more commercially valuable. Chowdhury adds that Flexion will need to work closely with hardware manufacturers to succeed and will face fierce competition. But without the ability to program humanoids in the way Flexion demonstrates, he says, “there isn’t really a market here.”