The robot, which looks like an orange mechanical ostrich, has knees to help it bend, just like the bird. It also has no cameras or sensors and is essentially “blind” to the world around it.
What makes Cassie extra special is the fact that it is the first bipedal robot to utilize deep learning to train itself. Previously, Cassie clocked just over 50 minutes for running 3.1 miles in 2021. The learning system within the droid allows it to measure its gait as it runs over different types of terrain so that it can adjust accordingly.
In preparation for the sprint, Cassie was installed with a year’s worth of training condensed to a week via a process known as parallelization, which allowed the bipedal mechanical runner to experience multiple training regiments simultaneously.
Ultimately the main learning point from the experiment is how to control robots better. Machine learning has been around for a while, but its use has rarely reached an end where a bot can study and adjust to its surroundings.