Arizona State University (ASU) researchers have recently shown how machine learning can be used by an autonomous robot to learn how to play basketball.
Their Robot named “Sun Devil – RX” repeatedly shot a ball to the hoop and learned how to refine his motor skills.
Sun Devil started from scratch. It didn’t have any kind of idea what kind of movement would allow it to get the ball into the hoop. The only requirement the research team gave it was the ball should be in the center of the hoop.
The robot tried out different movements in the beginning, some of which weren’t dynamic. Bit by bit it figured out how to improve its movements to get the ball closer and closer to the hoop. Automatically, it discovered that sinking the ball actually required a dynamic motion, and it discovered that motion.
The key was that the robot did all of this as quickly as possible, with a minimum of trials.
A novel reinforcement learning algorithm “Sparse Latent Space Policy Search” allows the robot to learn the task within only about 2 hours. The robot repeatedly throws the ball and receives a reward based on the distance of the ball to the center of the hoop.
