Google-DeepMind’s Barkour project is pushing the boundaries of robotics with robots that can match or even surpass the agility of real dogs. The good news is that the hardware is open source.
Meet Barky and his friends, the agile four-legged robots designed to help DeepMind learn “Animal-level agility“. But don’t just watch the video and read the document, you can download everything you need to replicate the hardware from the Google-Deepmind Barkour Robot repository on GitHub. The repository contains the design assets – CAD, PCBA designs, assembly instructions, etc. – you need to build your own Barky, which is licensed under an Apache 2 license. However, while core software (firmware and low-level code) is provided, no higher-level AI code is provided – although developing your own code has been made easier.
Although we’ve seen successive generations of robotic quadrupeds over the years, there have been few benchmark tests to compare the progress of these robots. To change that, starting in 2022, Google DeepMind researchers developed an agility course they called “Barkour.” Inspired by real-life dog agility competitions, it includes obstacles such as a break table, slalom poles, an A-frame climbing course, and a jump, and uses a time-based scoring mechanism.
To set a solid baseline, researchers from Google and Deepmind developed two methods to achieve the benchmark. In the first approach, they trained specific locomotion skills using reinforcement learning methods and combined them with a high-level navigation controller. In the second approach, they distilled the specific skills into a Transformer-based general locomotion policy that can handle different types of terrain and adjust the robot’s gait based on the perceived environment and robot states.
They then tried it out with a specially built four-legged robot called “Barky,” which could complete the course at half the speed of a dog – as demonstrated by training real dogs to complete the course:
Detailed information can be found in the report “Barkour: Benchmarking Animal-level Agility with Quadruped Robots” and summarized in this video:
Barkour vb is another new four-legged addition to the MuJoCo menagerie. It is a collection of high-quality, open-source robot models for the MuJoCo physics engine, curated by Google DeepMind, where it fits into a number of four-legged simulations, including Unitree Go, Anybotics ANYmal, and Boston Dynamics Spot.
More information
Barkour on GitHub
MuJoCo Menagerie on Github
“Barkour: Benchmarking animal-level agility with quadrupedal robots” on arxiv
Related articles
Four-legged parkour
ANYmale evolution and locomotion
A story of three (robot) dogs
Should we be wary of the Unitree Go2?
To be notified of new articles on I Programmer, sign up for our weekly newsletter, subscribe to the RSS feed and follow us on Twitter, Facebook or LinkedIn.
Comments
or send your comment by email to: [email protected]