Researchers from Oregon State University have created a bipedal robot called Cassie. Recently, Cassie traveled five kilometers, completing the entire journey in just over 53 minutes, making history in the process. Cassie was developed with direction from robotics Professor Jonathan Hurst with a 16 month, $1 million grant from the Advanced Research Projects Agency of the US Department of Defense.
Since Cassie was introduced at the school in 2017, students have been exploring various machine learning options for the robot. Cassie was the first bipedal robot to use machine learning to control her running gait on outdoor terrain. The robot completed the 5K on the Oregon State campus untethered and using a single battery charge.
Cassie is designed with knees that bend like an ostrich, and the robot taught itself to run with a technique known as a deep reinforcement learning algorithm. Running is complicated for robots because it requires dynamic balancing, which is the ability to maintain balance while switching positions or being in motion. Cassie has learned to make infinite subtle adjustments to stay upright while moving.
Researchers say that Cassie is very efficient because of how it was designed and built. Deep reinforcement learning is a powerful AI method opening up skills like running, skipping, and walking up and down stairs. Hurst says that robots will be common out in the real world one day, being as frequently seen as automobiles are today. The limiting factor behind robots so far has been the science and understanding of legged locomotion.
Researchers at Oregon State made multiple breakthroughs in that department. During the 5K, Cassie’s total time was 53 minutes and three seconds. The total time included about six and a half minutes of resets following two falls. One of the falls happened because of an overheated computer and the other because the robot was asked to turn at high speed. Cassie is also adept at walking up and down stairs.