A new study from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) demonstrates a significant advancement in robotic dexterity. Researchers have developed a system that allows a robot equipped with a two-fingered gripper to manipulate over 2,000 diverse objects, from mugs and plates to items like shoes and toys, with high reliability. The system, named ‘Dexterity Network’ or ‘Dex-Net’, uses a neural network trained on a vast simulated dataset of 6.7 million grasps. This enables the robot to adapt its grip in real-time based on visual input from a camera, successfully picking up and reorienting objects it has never encountered before. The research marks progress toward more general-purpose robots capable of functioning in unstructured environments like warehouses and homes. For the full details, read the article at https://technologyreview.com/2023/10/26/1082449/robot-gripper-learns-to-pick-up-anything.