A humanoid robot developed by Flexion Robotics, a Swiss startup founded by former Nvidia engineers, has demonstrated the ability to perform complex office intern tasks autonomously. The system uses an artificial intelligence model that learns from human videos and relies on simulation training combined with reinforcement learning. The result is a robot capable of opening doors, climbing stairs, and using elevators without human intervention.
Simulation training enables transferable real-world skills
Flexion's success lies in training the robot in simulated environments, where it learns individual skills such as walking, grasping objects, and maintaining balance. A master AI algorithm then determines how to combine these skills to perform a complex task. For instance, the robot was commanded to retrieve a snack package from the mailroom, using stairs and the elevator, and then unpack the items into a drawer in the break area. All operations were conducted without direct human input, contrasting with many other systems that rely on teleoperation.
Sponsored Protocol
Reinforcement learning as the secret ingredient
According to Nikita Rudin, CEO and cofounder of Flexion, the key element is extensive use of reinforcement learning, which allows the software to improve through trial and error. Every layer of the system, from the master AI model to motor control, adopts this technique. This approach makes the robot more adaptable to new environments compared to traditional methods. George Chowdhury, an analyst at ABI Research, highlights that the market for robot foundation models could be worth $150 billion by 2036, underscoring the importance of these developments.
Collaborations and future challenges for humanoid robots
Flexion is collaborating with several robotics companies, and its software is designed to work on different humanoid platforms. This could boost the commercial value of the technology. However, Chowdhury warns that the startup will face stiff competition and must work closely with hardware manufacturers to succeed. Without the ability to program humanoids as Flexion demonstrates, the office robot market would be very limited.
Sponsored Protocol
The original Wired article notes that the innovation lies not so much in the robot itself but in the AI models that support it. While leaders like Elon Musk and Jensen Huang predict a huge economic impact from humanoids, Flexion shows that the real challenge is artificial intelligence. For more on data security implications in AI systems, read the article on Google Warns Search Data Could Be Hacked Due to New EU Rules. Additionally, for an overview of reinforcement learning, check the Wikipedia entry.
Source: https://www.wired.com/story/this-robot-is-going-to-replace-your-interns-flexion
