It assesses what would be the best point of contact based on its previous experience with lifting similar objects. The system, called Dex-Net, is being developed at UC Berkeley. With its left hand, the robot can grab objects and with its right hand, it can suck objects up. Using deep learning, the robot can decide which of the two options has a higher chance of succeeding. In a recent experiment, the system’s current version was able to effectively pick up 64 items in less than 15 minutes!