New Robot Trick: Shake an Object to Learn What's Inside

Have you ever shaken a wrapped present to guess what's inside? Researchers at arXiv just taught a robot to do the same thing. Their soft robotic arm, built with an origami-style folding design, grabs an object and gives it a quick shake. A camera or small built-in sensors then watch how the object wobbles and settles. From that wobble, the robot works out where the object's weight is concentrated.

That's a real problem for today's robots. A sealed container looks identical whether it's full or empty, and cameras simply can't see the difference. But hidden weight changes how an object moves. The team leans on a clever idea called physical reservoir computing (using a robot's own flexible body as a natural calculator). Because the arm always shakes the same way, any difference in the wobble must come from the object itself. A simple math step then translates that wobble into a clear answer.

The arm completed three challenges, each harder than the last. First, it figured out which direction the object's balance point leaned. Second, it measured how far that balance point sat from where it was gripping. Third, it used that information to re-grip the object more securely. Better still, the team found that summarizing the wobble over time — rather than measuring it moment by moment — improved accuracy.

Why should you care? Robots that understand hidden weight can carry a coffee cup without spilling, stack boxes without dropping them, or sort packages without crushing what's inside. That means fewer broken deliveries and safer help in warehouses, hospitals, and eventually your kitchen. The catch: it still needs a good shake, so it won't work for fragile or tightly packed items — and it's lab-tested, not yet in real products.