Robots can do many things right now, but if there is one thing that robots have yet to learn, it would be how to open a door. That’s right, something that we probably take for granted is something that robots have difficulty learning, but that has changed, thanks to students at the University of Cincinnati.
In the past, researchers tried to address this problem by scanning an entire room to make a 3D model so that robots will know where the door is. However, the problem with this method is that it’s slow and that it’s not very practical since you would need to scan rooms and buildings one by one.
Also, the problem is that different doors have different types of handles that require different levels of force, so short of robots yanking the handles off, it is a challenge trying to get robots to learn how to open various types of doors. What the researchers have done is essentially put the robot through a process of trial and error combined with machine learning.
While this definitely takes a while for the robot to get it right, the upside is that it will eventually learn what to do. The researchers are currently using a digital simulation to help the robot prepare for the task, but according to the researchers, digital simulations are usually only 60-70% successful in the real world. They anticipate that it might take a year to help bridge the gap for them to perfect the system.