Robotic hands can play drums and even twirl objects with aplomb, but they're still poor at picking up unfamiliar objects. That's why UC Berkeley's DexNet 2.0 bot is so impressive -- using deep learning, it can successfully grasp random, real-world objects 99 percent of the time. What's more, the tech, developed with the help of Amazon, Google and Toyota, is far enough along that it could be put to work in manufacturing and supply chains in the near future. Researchers trained the DexNet 2.0 deep learning system using a vast library of 3D shapes and suitable grasp positions to match those objects. Using virtual, rather than real objects made it possible to train the AI much more quickly.