Google DeepMind researchers have just developed a standardized test to assess the performance of four-legged robots. Inspired by canine competitions, they designed an obstacle course that bears the name of… Barkour.
You may also be interested
[EN VIDÉO] When the Spot robot reproduces a choreography… The Boston Dynamics Spot robot reproduces the choreography of the music video “Start Me Up…
Robots have evolved a lot in recent years, gaining in speed and agility. One of the most popular shapes is the quadruped, or robot dog, like Boston Dynamics’ Spot. However, how do you compare different models when there is no standard to measure their performance?
To solve this problem, researchers from Google DeepMind decided to create an obstacle course intended to serve as a standard for measuring the performance of robots. Baptized Barkour, contraction of barking (barking in English) and parkourit is inspired by agility courses for dogs.
Robots half as fast as dogs
The course is part of a square of 5 meters on each side, and contains start and finish tables, a slalom, a palisade (or A-frame) with ascent and descent, and a long jump. They first tested Barkour with “dooglers”, the dogs of Google employees, who managed to complete the course in 10 seconds. The robot dogs used in the test took twice as long.
Researchers then score performance between assigning an agility score between 0 and 1. It is calculated by taking the maximum score, which is 1, and subtracting a penalty of 0.1 for each obstacle failure, as well as for the time spent beyond 10 seconds divided by 100, i.e. 0.1 for a robot that takes 20 seconds in total.
The researchers hope that Barkour can be used as a standardized test to compare different robot models, or to evaluate the performance of the same robot with different sensors, in order to eventually achieve animal-level agility.