Sign In

Communications of the ACM

ACM TechNews

Dog Training Methods Help JHU Teach Robots to Learn New Tricks

View as: Print Mobile App Share:
Johns Hopkins University Ph.D. student Andrew Hundt, with block-stacking robot Spot.

Johns Hopkins University researchers are using dog-training methods to help a robot teach itself new tasks quickly.

Credit: Will Kirk/Johns Hopkins University

Computer scientists at Johns Hopkins University used positive reinforcement to show a robot how to teach itself new tasks quickly.

The training technique, often used to change a dog's behavior, helped the robot improve its skills in a matter of days, rather than the month it previously would have taken.

The researchers used a reward system to help the robot learn from its mistakes.

When it came to stacking blocks, the robot quickly determined it would earn high points for correct behaviors and zero points for failing to grasp a block or knocking over a stack.

Said Johns Hopkins' Andrew Hundt, "The robot wants the higher score. It quickly learns the right behavior to get the best reward.”

From Johns Hopkins Hub
View Full Article


Abstracts Copyright © 2020 SmithBucklin, Washington, DC, USA


No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account