Robot Learns an Object's Weight and Size Before Giving it a Swing
MIT CSAIL's SwingBot is a robot that's able to learn the physical features of an held object through tactile exploration.
As humans, we take for granted being able to perform the most mundane tasks, such as handling objects of different sizes and weights, without thinking about how we perform the task. Think of it like picking up a suitcase that looks heavy, but in reality, is empty. We can adjust on the fly just by lifting the object, but that's not an easy task for robots. The reason we can pick up and handle objects we've never seen before lies in the fingertips, which can garner other information beyond size and weight, including texture, friction, and shape.
On the other side of the object manipulation coin, robots don't even possess the tactile skills of toddlers, so getting them to handle a single object is easy, and handling different objects of multiple sizes and weights is an incredibly complex task. That said, a team from MIT's CSAIL (Computer Science and Artificial Intelligence Lab) is working on that problem. They have developed a robotic arm with enough of a sense of touch that it can use a pair of grippers to pick up objects, estimate its size and weight, and swing it into nearly any desired pose.
The SwingBot platform is outfitted with tactile GelSight sensors in its fingers, and after picking up an object, it judges the weight and friction of the object by shaking it between its fingers. It then uses that information to plan the swing's timing and trajectory and pose at an error of only 17-degrees. CSAIL engineers state that the SwingBot could be useful for industries looking for more cost-effective and efficient tactile robots.