Watch: Robot Teaches itself How to Shoot Hoops in 2 Hours
The Sun Devil two-armed robot used a custom version of reinforcement learning to learn how to shoot hoops in two hours. Its creators say the learning technique can be applied to other areas of robotics.
Arizona State University (ASU) has been searching for its next men’s basketball star since James Harden left and was selected third overall in the 2009 NBA Draft. Well, it looks like the Sun Devils next star is being home grown in their robotics lab.
Called Sun Devil, this two-armed robot used a custom version of reinforcement learning to learn how to shoot hoops in two hours. Created by Heni Ben Amor, an assistant professor of computer science at ASU’s Ira A. Fulton Schools of Engineering, the algorithm is called “sparse latent space policy search,” and it enables a robot to first understand the coordination between its different joints, parts and movements. The robot then gradually eliminates unsuccessful solutions to arrive at a successful one.
“In a sense, this algorithm is linked to how humans learn - this project is not making a biological statement, it simply mirrors how we approach a problem,” says Ben Amor. “We innately understand the relationship between our different joints and synergistic movements, but this is something robots need to learn.”
Ben Amor says his approach is a faster-paced version of reinforcement learning, which “has its limitations, because it can take thousands, perhaps millions of trials for a robot to learn exactly what it needs to do to accomplish a task.”
Sun Devil’s creators used a two-arm design because it addresses other issues with machine learning and robot coordination. “Many robot coordination challenges simply employ a single arm, such as playing table tennis or cup-in-ball.”
Sun Devil also tackles dynamic movement, Ben Amor says. Shooting a basketball can’t be accomplished in starts and stops; it requires a burst of dynamic motion, which is something that eludes many learning robots.
“It requires a robot to dynamically apply force at the right time, straying away from the ‘divide and conquer’ approach common in computer science and machine learning,” says Ben Amor.