The bipedal humanoids may, in fact, be coming — but the quadrupeds are already here. They’re in labs, doing inspections in power plants and refineries, playing soccer and are even — much to the concern of many — becoming cops.
Boston Dynamics’ Spot is easily the most instantly recognizable of the bunch, but plenty of startups and research institutions have put their own spin on the category. Heck, even Xiaomi made one for some reason. While the purveyors of bipeds look to prove out their work, quadrupeds are getting the job done.
The team at Google’s DeepMind (which recently absorbed a large chunk of Alphabet’s beleaguered Everyday Robots team) just issued a research paper outlining a potential benchmarking system to quantify the performance of these machines. With a name like “Barkour,” one has to wonder whether the department worked backward from the title.
Google Research points to the various impressive feats accomplished by quadrupeds over the years, from hiking up mountains, running and jumping (“flipping is much easier than walking,” as an MIT professor once told me), but there hasn’t really been a baseline for determining system efficacy.
Given that these machines are inspired by animals, the research team determined that real animals would provide the best performance analog for their robotic counterparts. That meant setting up an obstacle course in the lab and having a dog run it — check out the tenacious little wiener above. The course was comprised of four obstacles in a 5×5 meter area, which it notes is denser than the dog shows that inspired it.
Performance is rated on a scale of 0 to 1 — a simple binary to determine whether the robot can successfully cross the space in the 10 or so seconds it takes for a similarly sized dog to do so. Various penalties are for slow speeds and either skipping or failing obstacles on the course. Google concludes:
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, and that our learning-based method for solving the benchmark provides a quadruped robot with a single low-level policy that can perform a variety of agile low-level skills.
The org says that Barkour has proven an effective benchmark even in the face of the inevitable unexpected event and hardware issues. The robot dog used in the trial was able to stand back up and return to the starting line on its own in the case of failure.
Google DeepMind introduces Barkour, a benchmark for quadrupedal robots by Brian Heater originally published on TechCrunch