Most of the robots being manufactured today are aimed at rescue and exploration tasks and have different cameras as a vision that allows them to read the terrain so as not to commit a false step.

However, these camera sensors are very sensitive, and a minor bump or accident could render them completely inoperative, making the robot no longer able to unwrap itself.

That is why now researchers from Oregon State University have made a bipedal robot without visual supportable to go up and downstairs. This robot, named Cassie from Agility Robotics, has been previously trained in a virtual simulator.

To achieve their goal, the researchers used a technique called ” real simulation reinforcement learning ” to establish how the robot will walk. First, they did the hit-and-miss phase under a simulator so as not to spoil the robot, and then they virtually taught the robot how to handle several situations, including climbing stairs or all kinds of terrain.

When the robot passed the simulated training they took it around the university campus to going up and downstairs and different types of terrain. After testing, it was able to successfully navigate even curbs, logs, and uneven terrain.

However, on the stairs, the researchers did 10 trials going up and another 10 trials going down the stairs, with 80% and 100% efficiency, respectively.

One of the main difficulties faced by the research team was when the robot was going too fast or too slow, which caused it to fail in these types of operations up and down stairs or terrain.