Quadruped robots have already made their mark, while the arrival of bipedal humanoids remains on the horizon. These robotic quadrupeds can be found in laboratories, carrying out inspections in power plants and refineries, engaging in soccer matches, and even taking on roles as law enforcement officers—raising some concerns along the way. [1]

Figure 1. Barkour.

Figure 1 shows a dog jumping. While Boston Dynamics' Spot stands out as the most recognizable quadruped robot, numerous startups and research institutions have also made their contributions to this field. Even unexpected players like Xiaomi have dabbled in quadruped robotics. While bipedal robot developers continue to refine their creations, quadrupeds are already demonstrating their practical capabilities. [1]

Recently, the DeepMind team at Google (which absorbed a significant portion of Alphabet's Everyday Robots team) published a research paper outlining a potential benchmarking system to evaluate the performance of these quadruped machines. It's interesting to note the name they chose for this system, "Barkour," which leads one to wonder if the team started with the catchy title and developed the concept from there. [1]

Google Research highlights the remarkable achievements of quadruped robots, such as scaling mountains, running, and even performing jumps (as an MIT professor once mentioned, "flipping is much easier than walking"). However, there has been a lack of a standardized baseline to assess the effectiveness of these robotic systems. [1]

To establish a performance benchmark for quadruped robots, the research team recognized that real animals would serve as the most appropriate performance analog. To achieve this, they set up an obstacle course in their lab and had a dog run through it, showcasing its determination and agility. The course consisted of four obstacles arranged in a 5x5 meter area, denser than typical dog shows that served as inspiration. [1]

The performance of both the dog and the robot is assessed using a scale from 0 to 1, representing whether the robot can successfully navigate the course within approximately 10 seconds, comparable to a dog of similar size. Penalties are imposed for slow speeds and for either skipping or failing obstacles. Google's conclusion emphasizes the importance of developing a benchmark for legged robotics to gauge progress towards achieving animal-level agility. The study demonstrates that Barkour serves as a challenging benchmark that can be easily customized, and their learning-based approach enables a quadruped robot to perform a range of agile low-level skills using a single low-level policy. [1]

The research team also notes that Barkour has proven to be an effective benchmark even when faced with unexpected events and hardware issues. In the event of failure, the robot dog used in the trial was capable of getting back up and returning to the starting line autonomously. [1]

References:

1. https://techcrunch.com/2023/05/30/google-deepmind-introduces-barkour-a-benchmark-for-quadrupedal-robots/

Cite this article:

Hana M (2023), Roaming the Wild: Barkour Showdown - Quadruped Robots vs. Animal Agility, AnaTechmaz, pp.275