Skip to content

Google DeepMind Unveils 'Barkour': A New Benchmark for quadrupedal robots

Google's DeepMind introduces "Barkour," a unique benchmark system designed to evaluate and quantify the performance of four-legged robots, paving the way for advanced developments in the field of robotics.

Google DeepMind's 'Barkour': A New Benchmark for Quadrupedal Robots

The future of robotics has quite a bit of paw-ssibility! While bipedal robots continue to evolve, their four-legged counterparts are already on the move, performing tasks in labs, industrial environments, and even law enforcement. Among these quadrupeds, Boston Dynamics' Spot may be the most famous, but the field is rapidly growing with entries from numerous startups, research institutions, and even tech giant Xiaomi.

Google's DeepMind, which recently took over a significant part of Alphabet's Everyday Robots team, has just launched a unique initiative: Barkour, a benchmarking system aimed at quantifying the performance of quadrupedal robots.

With Barkour, the DeepMind team hopes to create a uniform standard to measure the capabilities of four-legged robots, given their impressive feats over the years, such as mountain climbing, running, and even flipping. According to an MIT professor, "flipping is much easier than walking" for these robots!

Google DeepMind's 'Barkour': A New Benchmark for Quadrupedal Robots

Inspired by the natural agility and skill of real animals, the DeepMind researchers set up a specially-designed obstacle course in their lab. A canine participant (a determined little wiener dog) navigated a series of obstacles in a 5x5 meter area – a denser challenge than what one might see in traditional dog shows.

In Barkour, a robot's performance is rated on a binary scale of 0 to 1, based on its ability to traverse the course in roughly the same time as a similarly sized dog (about 10 seconds). Slow speeds, skipped obstacles, or failure to complete obstacles incur penalties.

Google DeepMind asserts, "We believe that developing a benchmark for legged robotics is an important first step in quantifying progress toward animal-level agility... Our findings demonstrate that Barkour is a challenging benchmark that can be easily customized."

Moreover, DeepMind has confirmed Barkour's effectiveness even in the face of unforeseen challenges and hardware issues. The robot dog used in the trials could independently recover from a fall and return to the starting line, proving its resilience and readiness to face the real world.