Blind Reinforcement Learning on Stairs

A team of researchers at Oregon State University has used one of our Cassie robots to demonstrate a robust learned gait that allows for traversal of stairs and other obstacles without any visual knowledge of the environment. Training first in simulation provides for a large training set with minimal risk to hardware. This remarkable result provides strong evidence for our view that successful locomotion is first and foremost a dynamics problem, and not a perception one. Many of the students involved in this project are now Agility employees and work continues with Digit... stay tuned. Reposted with permission from the researchers.