Oregon State University's Cassie, the bipedal robot, has made history by completing a 5-kilometer (3-mile) course in 53 minutes. Researchers said that Cassie is the first-ever bipedal robot that ran on outdoor terrain.
She ran the course untethered and powered by a single battery charge. Her record is a significant advancement for the entire robotics community. Cassie was produced by OSU spin-off Agility Robotics, which campaigns for the commercial use of bipedal robots.
Cassie the Bipedal Robot
According to OSU's news website, Cassie, the bipedal robot, was developed under the direction of robotics professor Jonathan Hurst, was funded by the U.S. Department of Defense's Advanced Research Projects Agency, and was launched in 2017. Since then, OSU researchers have been exploring machine learning to improve the robot.
Cassie is also the same robot that blindly navigated a set of stairs, as reported by Engadget in May. OSU engineers trained Cassie in a simulator to enable it to maneuver itself in going up and down the stairs without the use of cameras of LiDAR.
The team said that they used the same technique in teaching Cassie to run, which enabled her to do so without a tether or without seeing. It learned to shift its balance as it runs the course, as well as make subtle adjustments to accomplish the feat.
"Deep reinforcement learning is a powerful method in AI that opens up skills like running, skipping, and walking up and downstairs," Engadget quoted OSU undergrad Yesh Godse from Dynamic Robotics Laboratory.
Researchers said that the technique they used in teaching Cassie to navigate the stairs is called sim-to-real Reinforcement Learning (RL) to enable it to walk, noting that the training involved many falls and crashes. But the use of a simulator helped them avoid breaking the robot.
A Giant Leap for All Robots
Cassie's achievement made her the very first bipedal robot to use machine learning to run outdoors. She ran a 5K course within 53 minutes, comparable to humans who can do the same feat between 30-40 minutes.
Hurst said in the university's news release that the students in Dynamic Robotics Laboratory have combined their expertise in biomechanics and existing robot control approaches with new machine learning tools to enable Cassie to move at an animal-like level.
According to Robb Report, Cassie's training was not perfect as the robot fell twice while running when the human at the helm took a sudden turn at the corner and when Cassie's computer overheated. Despite that, she still made history which is a major milestone for all robots.
Agility Robotics hopes to see bipedal robots integrate into human society soon, and they are even exploring commercial avenues where it can be helpful, such as in delivering packages. Its new model named Digit has been trialed by Ford as part of a study in autonomous delivery robots. Digit has two hands that are capable of carrying packages.
Hurst noted that perhaps it would not be too long before people see and interact more with robots in different places every day as they work alongside these machines that can help improve the quality of life.
Check out more news and information on Robotics in Science Times.