Headless Robot Cassie Learns to Climb Stairs With No Vision

Developed by Dynamic Robotics Laboratory, Cassie is a headless robot based on a built provided by Agility Robotics that only uses its two legs to move around. After walking it through fire a while ago, the team of researchers is back with more testing. This time, they successfully taught the robot to go up and down the stairs with no vision.

While robots have long been able to master stairs by completely relying on cameras, in certain conditions, such as dim lighting or foggy weather, these devices might stop functioning properly. To overcome these issues, Cassie was trained to navigate using only “proprioception,” – the ability of an organism to perceive the position and movements of its own body in space. That means no cameras or any other depth tools were used.

Working virtually with a method called “reinforcement learning” to simulate real-world scenarios, the team prepared the robot for the actual tests. Cassie was subjected to ten trials which involved going up and down the stairs. It achieved locomotion in climbing up eight times and in going down had a 100 percent success rate. On top of its climbing abilities, the robot was also tested on steep hills and unstructured terrain.

Dynamic Robotics Laboratory published a paper which details the whole process that went behind their extensive work and a video to showcase the successful results that they achieved. Although Cassie does fall two times, misses a few steps, and stumbles, what is truly impressive is that it manages to get where it needs with no vision to ease its job. Heck, even I can’t climb the stairs with my eyes closed without losing my footing.