Google DeepMind’s Barkour project is revolutionizing robotics by introducing robots that can mimic or even surpass the agility of real dogs. The hardware for these robots has now been open sourced. Quadruped robots like Barky have been specifically designed for DeepMind to study “animal-level agility”. Interested individuals can access all the necessary information to replicate the hardware from the Google-DeepMind Barkour Robot repository on GitHub, which includes design assets such as CAD files, PCBA designs, and assembly instructions. While the core software is provided, higher level AI code is not included, although it has been simplified for users to develop their own.
In response to the lack of benchmark tests for robot quadrupeds, Google DeepMind researchers created an agility course called “Barkour” in 2022. The course, inspired by real dog agility competitions, features obstacles like a pause table, weave poles, an A-frame climb, and a jump, with a time-based scoring system. To establish strong benchmarks, researchers developed two methods to tackle the challenges. The first method involves training specialist locomotion skills using reinforcement learning techniques and combining them with a high-level navigation controller. The second method distills these skills into a Transformer-based generalist locomotion policy capable of adapting to different terrains and adjusting the robot’s gait based on the environment and robot states. These methods were tested on a custom-built robot named “Barky”, which completed the course at half the speed of a real dog.
For more detailed information, readers can refer to the report “Barkour: Benchmarking Animal-level Agility with Quadruped Robots”, as well as watch a summarized video. Barkour vb is a recent addition to the MuJoCo Menagerie, a collection of open source robot models for the MuJoCo physics engine curated by Google DeepMind. It joins other quadruped simulations like Unitree Go, Anybotics ANYmal, and Boston Dynamics Spot. Additional resources and related articles are available on GitHub, arXiv, and I Programmer’s website for those interested in the Barkour project and advancements in robotics.