A benchmark for quadruped robots
Meet Barkour designed to test agile four legged robots like the Boston Dynamic Spot
In this article, we discuss Barkour, a new benchmark created by Google Deepmind research group, designed to evaluate quadruped robots such as the Boston Dynamic Spot. Inspired by dog agility competitions, Barkour is an obstacle course created on (5m x 5m) 25m^2 square grid which challenges the robot to make it in the quickest time from the start table to the end table while avoiding the obstacles. Once the robot makes it through, it is assigned a benchmark score between 0 and 1 depending on the number of obstacles it missed, and the extra time the robot took beyond the allocated time. The allocated time would depend on the size of the robot and the average speed that the robot can attain. Starting from the highest score of 1, there is a deduction of 0.1 for every obstacle that the robot misses, and a deduction of 0.01 for every second the robot takes beyond its allocated time. The following picture shows a high level view of the obstacle course.
