As achievements go, learning how to pick up objects doesn’t sound quite as impressive as twice beating the world Go champion – it is, after all, something the average toddler can do. But it’s the fact that the robots themselves figured out the best way to do it using neural networks that makes this notable.

A recent Google report spotted by TNW explains how the company let robot arms pick up a variety of different objects, using neural networks to learn by trial-and-error the best way to handle each. Some 800,000 goes later, the robots seemed to have it figured out pretty well …

NordVPN

One of the most exciting aspects of the proposed grasping method is the ability of the learning algorithm to discover unconventional and non-obvious grasping strategies. We observed, for example, that the system tended to adopt a different approach for grasping soft objects, as opposed to hard ones. For hard objects, the fingers must be placed on either side of the object for a successful grasp. However, soft objects can be grasped simply by pinching into the object, which is most easily accomplished by placing one finger into the middle, and the other to the side. We observed this strategy for objects such as paper tissues and sponges.

You can see this behavior at work in the video below. You can also see that the robots learned that sometimes it’s easier to adjust the position of the object on the ground before picking it up – again, something learned by the system itself rather than taught.

Given that Google’s AI systems can now outsmart one of the smartest guys on the planet, and seem pretty adept at manipulating objects, I’d just like to place it on record now that I for one welcome our new robot overlords …

About the Author

Ben Lovejoy's favorite gear