When humans fall, they tend to have a sense of self-preservation. They stick out an arm, or a leg—anything to avoid hitting the ground nose-first. Robots? Not so much. 

But researchers at the Georgia Institute of Technology are teaching robots to fall gracefully, saving time and money for roboticists whose prize research project might break its neck—or its motor—trying to perform a task that might seem simple to a human, like walking over uneven terrain. 

Even the fanciest, smartest robots fall down sometimes. Just watch this gleeful compilation of cutting-edge ‘bots crashing into the ground at the DARPA Robotics Challenge:

The challenge was originally launched in 2011, just after the nuclear disaster at Fukushima, as a way to encourage development of the kind of robots that could eventually replace humans working in highly dangerous areas—like repairing a downed nuclear reactor. In the places where these kinds of robots would be most useful, cleaning up and providing humanitarian aid after natural disasters, for instance, they’re also most likely to take a tumble over something unexpected. Far from the research lab, they need to be able to get back up again. 

The Georgia Tech algorithm allows a robot to calculate how to hit the ground with less force, so it doesn’t break itself. An accelerometer in the robot’s head and a motion-capture camera are the nervous system, in essence, giving it something akin to a human’s reflexes. Instead of falling however gravity takes it, the robot attempts to make more than one contact point with the ground, dissipating some of the energy of the fall. 

So far, the algorithm has only been tested on one robot, and in simulations with another, but given how many of the DARPA contest participants let gravity get the best of them, there’s plenty of test subjects to work with in the future. 

[h/t: MIT Technology Review]

Banner image screenshot via YouTube