Robots Can Now Take Apart Old Gadgets Just by Listening

Taking something apart is much harder for a robot than putting it together. A factory robot welding the same joint a million times is easy. But disassembling a used laptop or appliance means dozens of fiddly steps: find the screws, unclip the connectors, pull the right part in the right order. Every model is built differently, so robots can't just memorize a script.

This team's fix was to let the robot read the room — literally. They gave it plain-language instructions, like "disconnect the battery," and a vision system that matches those words to the actual part in front of the camera. The robot learns by imitation (watching humans do the task, then copying them) rather than being programmed step by step. The language acts like a label telling the robot which skill to use and where to aim it.

The results were a big jump. Their method succeeded 35 percentage points more often than the previous best robot approach, and 75 percentage points better than an earlier version that knew the task but not the language. Crucially, it worked across many different connector shapes and product layouts without anyone hand-labeling every part in the images — a slow, expensive step that usually holds this kind of research back.

The bigger picture: the world generates roughly 60 million tonnes of electronic waste a year, and most of it is still taken apart by hand or simply thrown away. Robots that can safely dismantle devices could make recycling and repair cheaper and faster, and keep valuable metals out of landfills. The catch is that this is still lab work — the robot performs in controlled settings with training data from humans, so don't expect a robot repair shop next week.