If a robot has pick up a workpiece that is in an unusual position on a conveyor belt, it will probably grasp beside it - unless it detects the deviating position and adjusts its gripping motion accordingly. As "Engadget" reports , a self-learning system capable of doing just that is being developed by scientists at the MIT Computer Science and Artificial Intelligence Laboratory. DON is the name of the creation and it stands for "Dense Object Nets".
The DON system apparently teaches itself to grasp the handle of a coffee cup and then hold the cup without spilling its contents while moving it. In the context of a production facility, for example, DON would be able to pick out a specific object from a stack of similar objects. The MIT researchers have published the details in a paper ; the code is also available on Github . There is also a video that shows DON's special abilities.