With the technological advances in robotics that have emerged in recent years, several feats have been achieved. One of them has been to make a four-legged robot with computer vision able to climb stairs without problems. However, replicating this same result in a blind bipedal robot has been a real challenge.
It was not until a group of researchers from Oregon State University achieved this feat with their bipedal robot named Cassie, which they trained in a simulator.
The reason behind this project is to enable robots to have the ability to move in a poorly lit environment or in conditions where sensors and cameras cannot perform effectively. In this way, the project aims for robots to support their motor skills on proprioception (body awareness).
To do this, the researchers used a technique known as symbolic reinforcement learning (RL) to determine how the robot will move. The researchers expressed that within the training for bipedal locomotion it is normal for many falls and crashes to occur, especially at the beginning.
That is why they opted for the use of a simulator in order to preserve the good condition of the robot. In this sense, the bipedal robot was instructed so that it could move virtually on different terrain such as stairs or flat terrain.
After the simulated training was finished, the researchers made the robot move around the university campus, climbing stairs and different types of terrain.
In the end the robot performed well, successfully handling curbs, logs and other uneven terrain that I had never seen before. When going up and down the stairs, the researchers subjected the robot to 10 ascent and 10 descent tests, in which it obtained an efficiency of 80 and 100 percent respectively. You can learn more about the development process of this robot HERE