The new robots learning to work in offices

Flexion Robotics, founded by a team of former Nvidia engineers, is developing humanoid robots—machines designed with heads, torsos, and limbs to mimic human form. Instead of using traditional programming where every move is manually typed out, the company is using a training method that helps the robots learn by observing and repeating actions. This allows them to handle tasks that would normally require a human hand to guide a machine through an ever-changing workplace.

Traditional industrial robots are like a calculator; they only do exactly what they are told based on rigid, pre-set rules. If the table moves an inch, the robot fails. Flexion’s approach is closer to an apprentice model. The robot is equipped with cameras that act as eyes. During training, a human performs a task—like picking up a stapler—while the robot records the camera footage of the hand movement and mirrors it with its own mechanical limbs. The robot’s internal system constantly compares its own simulated "view" of the task to the human's example. It tries the motion thousands of times, receiving a "score" for how closely its movement lines up with the human's. Eventually, the machine identifies the mathematical patterns that lead to success, allowing it to "see" a cup and know how to move its arm to grasp it, even if that specific cup has never been seen before.

The shift from 'pre-programmed' to 'learned' behavior is what makes these machines suddenly seem more capable. If robots can successfully mimic interns, it opens the door to machines that don't need a perfectly optimized factory floor to function. We are moving toward a reality where robots might assist with physical labor in ordinary offices. The core question is no longer whether a robot can lift a box, but whether it can navigate the messy, unscripted chaos of the human world without needing a human to write code for every single step it takes.