Robotic hand uses AI to specialize its grip for any object

It’s not easy to build a robot that can compete with that astounding piece of machinery. But the UK company Shadow has a solid contender. And while it’s robotic hands imitate nature, why keep nature’s limitations? Shadow’s Dexterous Hand “sees” the shape of the object approaching it to help it determine how to hold it.
The hand makes clever use of Microsoft’s Kinect off-the-shelf depth sensor to size up the object it needs to grasp using a 3D point cloud. If the hand has seen the object before, it calls up instructions for the most practical shape to adopt to grasp it. If it hasn’t seen the object before, it runs artificial intelligence algorithms to guess at the best shape. (If its guess is lousy, it will feed that information back into the system to make a better effort next time.)
Capable of 24 distinct movements, the hand can come up with an almost human array of positions to hold a Rubik’s cube, business card or light bulb. The finger lengths vary like a person’s too, giving the hand a more organic feel than many of its competitors.