UC Berkeley’s VelociRoACH robots are something like a decade old — the first one, DASH, was delivered at IROS 2009, back when IROS was small enough to fit into a Hyatt in St. Louis, Mo. The upgraded VelociRoACH showed up last year, and we’re still seeing it being used for innovative new research. The great thing about these little robots is that they’re inexpensive, easy to create (mostly cardboard), and even easier to adjust, so they’ve evolved promptly over the years with things like wings and winches and drone launchers.
 
A few years ago, the addition of a shell (which actual roaches have) plus a tail (which actual roaches gratefully do not have) allowed VelociRoACH to flip itself over if it ended up upside down. This performed really well, but it did add a little bit of problem and expenditure to the VelociRoACH design. Not situation or expenditure that you’d care about if you were just making one robot, or 10 robots, or even maybe a 100 robots, but the whole point of making super cheap little mobile robots like VelociRoACH is that you want to be able to churn out thousands of them, and then deploy them in massive swarms to (say) find people in rubble after an earthquake.
 
The latest version of VelociRoACH leverages the swarm idea to solve the flipped-over robot problem using nothing more than a slightly-redesigned shell. Instead of using a cockroach-like rounded robot, a square-fronted shell allows one robot to simply smash itself headfirst into another robot until it flips it over.
 
The design of the shell is a bit more confusing than just cutting up the ends off to make it square-ish instead of round-ish. The profile has to be circular rather than elliptical, and the addition of some rubberized pieces turns out to be necessary for the robot to flip over at all. It’s also an intentional compromise between size and righting efficiency. With a much larger shell, you could either take advantage of the Weeble effect, or design the shell to work like a Gömböc, helping the robot flip itself through clever geometry. But doing either of those things would hit the ability of the robot to skitter through small spaces.
 
The overall success rate of this robot-on-robot righting method is 87 percent, but it doesn’t really matter, because there’s no reason not to just keep trying over and over. It also doesn’t actually matter whether these robots have sensors on them that can discover other robots and localize them with sufficient quality to nudge them in correctly the right way on the first try, because if you know within a few tens of centimeters or so where the upside-down robot is, you can just keep running back and forth through that area until you run into it and flip it over successfully. This is the nice thing about cheap robot swarms—by embracing their lack of sophistication, it’s possible to find ways to make them as effective as robots that are more complicated and expensive.



This article is originally posted on Tronserve.com