In BriefThe team equipped a robot called Baxter with powerful deep learning capabilities, placed a table full of ordinary objects in front of it and then left it to learn, like a baby playing in a high chair
Baxter is a modern two-armed industrial robot designed to perform basic, repeatable tasks. In this experiment, Baxter was programmed to identify a single object, pick a specific point, and then attempt to grasp the object. Success was then determined by the robot’s camera and information relayed by the force sensors.
Over 700 hours, Baxter performed some 50,000 grasps on 150 different objects, each time learning whether the approach was successful or not. Objects included a TV remote and a variety of plastic toys. This reinforced learning system ultimately led Baxter to determine whether an incoming approach would be successful over 80% of the time, far surpassing other techniques.
This research could have an important impact on the way that robots interact with complex environments. Currently AI’s struggle to navigate and interpret cluttered and complex environments that humans can easily adapt to. At present, Baxter and its convolutional neural network have a long way to go before they can match the grasping skills of a young child. A key area of focus in the future will be determining how hard to grip a specific object without damaging it.