Does Our Robot Have Bird Legs?

We’re often asked if Digit’s legs were intentionally designed to mimic bird legs. If you’ve ever watched a heron or a crane walk through a field or an ostrich run, Digit’s legs look very similar. But there’s a major difference in how we arrived at our design.

Bird legs evolved over millions of years while Digit’s legs are the result of 20 years of bipedal locomotion research examining how people and animals walk, research that has been applied toward engineering robotic legs that excel at dynamic stability.

“Early on trying to capture the dynamics of how animals walk and run, we first looked to biomechanics,” said Jonathan Hurst, our co-founder and chief technology officer. “We looked at basic math models of how animals walk and run.”

The engineering behind Digit’s legs means it can go wherever we go. It can navigate a warehouse or office space, or even walk down a street or along a beach. Instead of focusing on mimicking nature, our engineering process honed in on optimizing stability and agility in movement that would allow Digit to do many different tasks. Consequently, the similarity is more of a coincidence than an intentional mimicry of mother nature.

The research that led to the development of Digit’s legs began back in 2009, in the Dynamic Robotics Lab at Oregon State University with Digit’s grandparent ATRIAS. The research team set out to build ATRIAS using a spring-mass model, essentially applying math modeling to determine how a human or animal can run or hop with speed. “The whole idea was to build a machine that’s as close to this model as possible, and then write really simple controllers,” Hurst said. “Apply them to the robot and see if we could get it to really work well. ATRIA was a success.”

Looking much like a box on a pair of stilts that could bend, ATRIAS was very different from Digit’s sleek, slender design. But ATRIAS paved the way for Digit’s functionality. As the first bipedal robot to reproduce the dynamics of the human walking gait, ATRIAS could walk and run over uneven and unsteady terrain, and it did all this without visual perception sensors.

There was a long way to go, however, toward perfecting its movements. “It was a great science demonstrator, but there were all sorts of things wrong with it,” Hurst said. “First of all, it can’t actually stand. It can’t turn.”

In 2013, Agility Robotics had spun off from the OSU lab into a private company to apply this knowledge to a commercial product, and a new iteration of ATRIAS was created – Cassie. 

“If you look closely at the Cassie design, there’s shadows of the ATRIAS design. Just move one of those linkages far in and really close, and you start to get what looks a lot like the Cassie leg,” Hurst said. “At that point, it looks like a bird leg, but we certainly didn’t set out to make it look like a bird leg.”

Birds are digitigrades, meaning they walk and run on their toes as opposed to flat feet like humans. It’s hard to make a one-to-one comparison between animals and robots or muscles and actuators, but the similarities between Cassie’s legs and bird legs are undeniable.

Cassie’s bipedal design was passed on to Digit, and a torso and manipulators were added. Unlike ATRIAS and Cassie, which were effectively blind, Digit has cameras and LIDAR, which it uses to perceive and navigate its environment. So if Digit is more like a digitigrade than a human, that means Digit’s knees are actually ankles, and its foot is actually a toe, just like for a bird. 

There are a lot of surprising benefits to this design. Digit’s legs can fold up behind its back for compact transportation or storage, and Digit can approach a shelving unit and squat straight down without the legs getting in the way. So far, we’re pleased with the design.

“When we have a physics-first engineering solution, and it ends up looking something like an animal, it’s a good indication that we’re doing it for the same reasons,” Hurst said. “It’s a hint, at least, that we’re going in the right direction.”

Previous
Previous

Exploring the Future of Robotics Through Art

Next
Next

Machine Learning and Artificial Intelligence for Robot Control