Algorithm teaches robots how to fall gracefully

If you have ever watched video from humanoid robot competitions, chances are you have seen a robot fall. Watching toddlers can tell you that it can be hard for humans to keep their balance, but we all have built in instincts to put our hands down to brace the fall and protect ourselves from injury. Robots don't have instincts, so they have to be taught how to fall gracefully and that is where a new robot algorithm comes in.

Researchers from Georgia Tech have designed a new algorithm that is aimed at teaching robots how to fall without seriously damaging themselves. Ph.D. grad student Sehoon Ha and Professor Karen Liu have developed the new algorithm to tell robots how to react to a wide variety of falls.

The algorithm teaches the robots how to take a single step to recover from a gentle nudge to a rolling motion to break a high-speed fall. The algorithm was validated using a physics simulation and tested using a BioloidGP humanoid robot.

Liu says that the team knew that the robot had the computational power to achieve a softer landing during a fall, but lacked the hardware to allow it to move fast enough to brace itself. The video shows the robot falling and using the algorithm to protect itself from damage.