Every year, the Gold Glove award is given to Major League Baseball players that exhibit superior fielding abilities. Now, it's time to meet the robot whose fielding percentage would be the envy of 18-time Gold Glove winner Greg Maddux. A robot developed by researchers at Ecole Polytechnique Federale de Lausanne (EPFL) in Switzerland adjusts on the fly to catch a variety of objects tossed in its direction. Empty and half-full bottles, hammers, tennis rackets and balls — of course — were all safely snatched out of midair by the 1.5-meter-long robotic arm. The key to the robots’ sticky fingers is its high-speed computing power: It can interpret and adjust to objects’ trajectories in less than five hundredths of second to make the catch.
Researchers didn’t push out lines of coded instructions to design a robot that can handle the knuckleball. Instead, they used a technique called “programming by demonstration,” which relies on humans to manually guide the arm as objects are tossed to it. Basically, they played an old-fashioned game of catch and taught the robot by manually moving it into the correct position to make each catch. A series of cameras then captured the movement of items thrown in its direction, and the robot created a model based on trajectories, speeds and rotation of various projectiles. After several human-guided catch attempts, the robot eventually “learned” how to catch an object by making real-time adjustments within milliseconds. They published their findings earlier this week in the journal IEEE Transactions on Robotics. Aude Billard, head of the Learning Algorithms and Systems Laboratory at EPFL where the arm was programmed, said the robotic arm could be incorporated into machines that help humans by catching them if they are in danger of falling or grabbing objects that may fall on them. The arm could also pluck space junk hovering in low-Earth orbit.