Boston Dynamic’s various bots have been amassing a series of superpowers. Atlas can backflip. SpotMini can open doors and fight back against interlopers. But they aren’t the only game in town. MIT’s Cheetah bot, which had already demonstrated its ability to hop hurdles mid-sprint, is now leaping up onto tables, and navigating completely blind.
The recent demonstration of the Cheetah 3 model demonstrates a few different improvements. First is the robot’s ability to navigate while effectively blind. Instead of relying on visual input from a camera or other sort of sensor system, the Cheetah 3 is designed to be able to feel its way around the environment, which could make uniquely suited to traverse unpredictable and/or pitch black spaces. In addition to this, much like its Boston Dynamics brethren, it’s learning how to put up with being shoved and pulled without immediately falling over.
But perhaps most impressive is its leaping capabilities. The Cheetah 3 is now able to rear back on its hind legs, leap into the air, and make a solid landing on a platform much taller than it is, though presumably this trick requires a little bit of visual input. All of these tricks lean on algorithms that pull information from the Cheetah’s legs 20 times every second to help it predict what is about to happen to it, and the best way for it to counteract any unexpected forces.
You know, like the desk you might throw at it while it’s chasing you down.
Source: MIT News
Previously Published by: Popular Mechanics USA