Robots Learn to Share Skills Across Different Designs

Teaching a robot a new task used to mean starting over every time you switched to a different model. Swiss researchers just changed that.

A team at the Swiss Federal Institute of Lausanne developed a system that lets robots with different designs learn from one human demonstration. When you teach one robot to pick up an object, other robots can now do it too, even if their joints move differently.

The secret isn't fancy AI. Instead, researchers mapped the physical limits of robotic arms to understand what movements each design can safely make.

Every robot has danger zones where joints can't bend or might move unpredictably. Engineers call these "singularities" because they cause the math controlling robot movement to break down, leading to sudden, unsafe jerks.

The team sorted three-joint robotic arms into six categories based on their shared physical constraints. They gave each robot what they call "kinematic intelligence," basically self-awareness of what it can and can't do safely.

When a movement pushes toward a danger zone, the system activates a "track cycle" that navigates around the problem. Some robots move horizontally around obstacles, others vertically, depending on their category.

To test it, researchers set up a mock assembly line with three commercial robots of varying flexibility. A human demonstrated pushing an object off a conveyor belt, picking it up, placing it on a workbench, and putting it in a basket.

All three robots completed the tasks successfully on the first try. Despite their different designs and movement capabilities, each one followed the human demonstration safely.

The Ripple Effect

This breakthrough could transform how factories adopt robotic automation. Right now, companies hesitate to upgrade robots because retraining costs time and money.

With skill-sharing technology, a warehouse could mix and match robots based on what's available or affordable. Train one, and the rest learn automatically.

The system works best in controlled environments like factories, not messy real-world settings with unpredictable obstacles. But for industrial settings where most commercial robots already work, it's a perfect fit.

The research, published in Science Robotics, opens doors for smaller manufacturers who couldn't previously justify the training costs of robotic systems. Now robots are becoming more practical, more affordable, and easier to deploy at scale.