Google's robotics lab now focuses on machine learning

Google has partnered with researchers from Princeton, Columbia and MIT to create TossingBot, which helps to learn how to pick up and toss various objects in the right containers. During his first rodeo, the mechanical arm did not know what to do with the pile of objects that was presented to him. After 14 hours of trial and error, analyzed using her cameras, she was finally able to throw the right object in the right container 85% of the time.

As the tech giant explains, programming a robot to grab and properly launch specific objects – a screwdriver, for example, could land in different ways, depending on where you hold it – is extremely difficult. Using machine learning, the robot will learn instead of experience, adapt to new scenarios and learn on the fly. This is the type of machine that would be useful in warehouses and distribution centers, such as those in Amazon or UPS.

In addition to TossingBot, Robotics at Google is working on a R2-D2 robot that uses machine learning to navigate in new spaces. Researchers also form a set of mechanical hands with fingers to manipulate objects by pushing, pulling and rotating them.