Oregon State University has a new track star in the making. Following its completion of a five-kilometer run last summer, Guinness World Records confirmed that Cassie’s 100-meter dash is the fastest ever completed by a bipedal robot. 

Coming in at 24.73 seconds, Cassie’s time is still about 15 seconds slower than the current human world record holder, Usain Bolt. While Cassie can’t yet beat humans, the feats achieved by Oregon State’s Dynamic Robotics and Artificial Intelligence Lab are a huge step in the development of humanoid robots.

Created in collaboration with Agility Robotics and Jonathan Hurst, Cassie was introduced to OSU in 2017. Modeled after large bipedal birds, Cassie learned how to walk just like the rest of us: through trial and error.

According to the Associate Head of Research and Professor of Computer Science Alan Fern, it took two and a half years to find the right controller architecture and algorithms needed to begin simulated trials. Trials are conducted this way for multiple reasons, including protecting the robot from possible fall damage and being able to conduct a year’s worth of attempts in one week of real time.

Engineering doctorate candidate Devin Crowley, who led the Guinness initiative for his master’s project, described how simulations allowed Cassie to learn how to walk so quickly. 

“The controllers that we use are fully trained after about a year of real-time experience,” Crowley said. “So if we were to train it on the actual hardware of the robot, it would take a year of experience ignoring repairs and set-up time. But since we run it in simulation, each experience can be experienced faster than 1x speed, but the bigger effect is that it’s able to practice with 50 versions of itself at once.” 

Besides being able to complete a year of simulation in only a week of real time, simulation also allows the robot to safely interact with a variety of terrains and environments.

“We randomized things like friction, so when the robot is learning in simulation it’s learning to be robust to all kinds of variations the world might have,” Fern said.   

According to Crowley, who controlled Cassie during the record-breaking run, this ability to adapt to its environment aided the robot during the attempt.

“My favorite part was seeing this one moment that’s in the video we cut together where it’s really listing off to the right,” Crowley said. “It looked like it’s about to fall, but it corrects itself.” 

OSU Dynamic Robotics has further aspiration for Cassie beyond running track. Soon, the robot will gain the sense of sight through mounted cameras, opening up many new possibilities.

“What vision allows you to do is you now can see the world and plan,” Fern said. “So for example, we have a project that we’re working on: ‘Stepping Stones.’ We want to put down stepping stones in front of Cassie where it can only step on certain things and have it be able to plan out the motion it needs.”

Fern added that the department also wants Cassie to learn how to find and go to places she’s ordered to, even if they are not in view. The professor compared what they want Cassie to do to the job of an Amazon delivery driver navigating a customer’s driveway in order to drop off a package. 

Using this AI technology with a model like Agility Robotics’ Digit, which comes fully equipped with a torso and arms, robots could have real-world applications in warehousing and delivery jobs. The thought of robots taking over jobs is a source of anxiety for some people, but Fern assures that the benefits greatly outweigh the negatives.

“I think a lot of people look at these robots and AI and usually a lot of times a science fiction, negative, dystopian future is what comes to mind,” Fern said. “But I think for every single negative thing you can think of for these technologies, there’s going to be ten positive things that influence the future.”