:format(webp)/cdn.vox-cdn.com/uploads/chorus_image/image/51684357/CIDSE_Heni_Ben_Amor_lab_JH_9319a.0.jpg)
Robotics researchers at Arizona State University have a invented a robot that taught itself to shoot hoops — just in time for the beginning of basketball season.
Using what’s called reinforcement learning algorithms, the two-armed droid learned to sink a ball through a net in only two hours. They’re calling it Sun Devil.
Its hands are made of prongs, which are capable of gripping, releasing and aiming a ball. To make a basket, the robot had to learn to move dynamically.
At first, Sun Devil dropped the ball, as its arms had to learn to move in synchronized time. Soon, it was able to throw, albeit far too low to make the hoop. But eventually, after failing and learning from its mistakes, the robot was able to land a basket.
Writing code for this type of robot, with two synchronized arms that aim and release an object, would typically take even an experienced programmer days.
Reinforcement learning allows robots to acquire new skills from practice. It’s how Google’s Alpha Go computer learned to defeat the human Go champion, Lee Sedol, earlier this year. Even huge industrial robots are starting to teach themselves to take on new tasks without requiring an engineer to reprogram them.
While Sun Devil is pretty good at shooting hoops, for now the robot is stuck in one place and lives in a lab.
Watch it fail and learn.
This article originally appeared on Recode.net.
