As you can see in the next video, this robot has learned to support himself with his hands reflexively every time he falls.
The design of this humanoid robot has been carried out by Duke University’s engineering professor Kris Hauser.
Hauser has programmed the software to focus only on the joints of the robot’s hip and shoulder. As Hauser explains:
If you push a person to a wall or a railing, you can use that surface to stand with your hands. We want robots to be able to do the same. We believe that we are the only research group that works for a robot to choose dynamically where to place their hands to avoid falls.
As long as the robot does not twist when it falls, this creates only three angles that the stabilization algorithm must take into account: the foot to the hip, the hip to the shoulder and the shoulder to the hand. In this way it identifies the surfaces near the range and then quickly calculates the best combination of angles. The algorithm takes its best estimate and then optimizes it progressively using a method called direct triggering.
Fortunately by the end of the year we should be experimenting with the robot really working on a live obstacle course. Then we will try to make the robot make a dynamic map of what is around him and reason about how to protect himself from falling into arbitrary environments.