If you’re walking along and stumble over something, your first instinct is to reach out and try to catch yourself from falling. If nothing is reachable, you fall and hurt yourself. Even if you’re just on flat ground, that can easily result in a broken wrist. Robots are also susceptible to damage from a fall, but most aren’t capable of even trying to catch themselves. That’s why roboticist Kris Hauser has developed algorithms to help bipedal robots catch themselves just like humans do.
Hauser is an associate professor of electrical and computer engineering, and of mechanical engineering and materials science at Duke University’s Pratt School of Engineering. He’s put all of that knowledge into unique algorithm that’s designed to give bipedal robots the ability to use their environment to stop themselves mid-fall. Give his small humanoid robot a shove, and it will reach out to a wall or nearby object in order to keep from falling over completely.
That’s possible thanks to Hauser’s efficient three-link model and algorithm for processing potential actions. Because the robot’s Raspberry Pi brain needs to process multiple outcomes in the short time while it’s falling, simulating all of the joint movements would be too slow. Instead, Hauser’s model just takes into account the angle between the leg and torso, and the angle between the torso and arm. With that simplified geometry, it’s able to quickly determine which part of the environment can be reached, and how best to reach them.