If you’ve ever seen herding dogs, like an Australian Shepherd or Border Collie, compete in agility competitions, you know how remarkably athletic they can be. That comes from a combination of natural evolution and intentional breeding as working dogs. But that kind of agility is difficult to replicate artificially in robotics — mostly for mechanical reasons. That’s why a team from the Stanford Student Robotics Club’s Extreme Mobility team used clever programming to make their Doggo robot perform more like a real dog.
Animals have evolved over millions of years to adapt to their environments. Dogs in particular are unique, because they were the first animal to be domesticated — a process that occurred sometime between 15,000 and 40,000 years ago. Their agility comes from their muscular and skeletal structure, which combines power with fast-twitch muscles, while ligaments, tendons, and bones are able to absorb shocks with ease. Those kinds of biological systems are extremely complex, and difficult to construct using artificial materials like metal and rubber.
Instead of even trying to recreate that biology mechanically, the Stanford team turned to sophisticated programming. The Doggo robot is mostly rigid, and each of its four legs are four-bar linkages driven by belt pulleys. When a real dog jumps up to grab a Frisbee and then lands, the muscles and tendons in its legs act like springs to absorb the shock. Doggo takes a different approach. It’s sensors and motor controller are updated 8,000 times per second, and perform like virtual springs by turning at the appropriate rate to allow the robot to land smoothly. Because those motors are powerful and Doggo is lightweight, it can even perform acrobatic feats like back flips.