Bipedal Robot Cassie Is an Official Guinness Record Holder in Sprinting

Cassie is a 2017 bipedal, headless robot designed by a team at Oregon State University College and built by Agility Robotics, and used as a platform to further study and develop AI and robotics. So far, it’s been put through the wringer and has performed admirably every time – an incredible feat, especially given the fact that it relies on just two legs for locomotion.

For its latest stunt, Cassie was able to run a sprint of 100 meters (328 feet) in 24.73 seconds without falling once. Like that time it was first able to climb stairs, Cassie is running blind, meaning it’s not relying on cameras or sensors, but only uses reinforcement learning methods to adapt to the environment. This recent challenge came about after Cassie was able to run 5 km (3.1 miles) untethered and on a single battery charge, which prompted researchers to see how far they could take the robot in terms of speed.

The test took place at the Whyte Track and Field Center on Oregon State’s campus and saw Cassie start running from a standing position and return to the initial position without falling. As the video below shows, the sprint included no falls, though it came close, at least on a couple of occasions. Cassie is the first bipedal robot to regulate a running gait on outdoor terrain using machine learning, researchers say, and this Guinness World Record will be a watershed moment in robotics.

For the record attempt, Cassie received a year’s worth of training in a simulation environment in the space of a single week, thanks to a computing technique known as parallelization, which allows for multiple calculations and processes to take place simultaneously. As artificial intelligence professor Alan Fern, who worked on Cassie, points out, the start and end of the sprint were the most challenging, but Cassie aced them.

“Starting and stopping in a standing position are more difficult than the running part, similar to how taking off and landing are harder than actually flying a plane,” Fern explains. “This 100-meter result was achieved by a deep collaboration between mechanical hardware design and advanced artificial intelligence for the control of that hardware.”

Oregon State robotics professor Jonathan Hurst says that Cassie’s new achievement shows that using learning policies for robot control renders improved performance over other control methods. Cassie might be the first bipedal robot to run, but it won’t be the last because it’s paved the way for real progress in the field, Hurst adds.