Cockroach-Like Robot to Help Explain Animal Movement

neutron_p writes "A cockroach-like robot named RHex is the starting point for a major project to understand animals' most distinguishing trait: how they move without falling over. Researchers from several universities will focus on RHex, a short, six-legged robot that scampers like a cockroach, as a working model of the principles they're seeking to uncover. By tweaking the robot and using it as a physical model, they hope to tease apart the complex neural and muscular networks in insects."

Comments

duh...

[niteice]

They don't fall over because they (usually) aren't missing a leg where one is needed for proper balance.

I work on this...

[feelyoda]

on the perception side...

While the ability of the bot to go over hard terrain is amazing, the point is that your relinquish direct control.

The basic problem in perception is dealing with the drastic motions.

The computer vision methods needed are quite complicated, requiring complimenting sensors like inertial measurement devices. Also extremely wide-angle cameras are excellent because things stay in view, but difficult because the pin-hole model fails.

Go here for some work that is now a bit dated, from a 180degree camera strapped to rhex:
http://www.frc.ri.cmu.edu/projects/buzzard/ rhex/

Re:I work on this...

[feelyoda]

"Could it be possible that a computational approach isn't necessarily the way to go?"

As Brooks showed, you can get very complicated behavior from reactive or even semi-stateful robots. BUT, I would question the scalability to something more application specific and useful.

For instance, imagine such a bot making a sandwich, and then cleaning your toilette...

Also, as some point, you're going to want to give the bot an order, like go from A to B to C then back. Rhex would be unable to do that currently without a very engineered environment, which goes against the entire point of such a bot which moves skillfully in all environments.

Adding a robust perception loop around the sense-response robot is the way to go, as far as I'm concerned.

$30M for more insect robots? Sounds like pork.

[Animats]

There's way too much work on insect-level locomotion. Brooks was doing this stuff twenty years ago, and took it about as far as it's going to go. Reactive systems reached their ceiling years ago.

We realy should be doing better than this. We should at least have Aibo-type robots running (or at least trotting) over real terrain by now. It's embarassing.

The trouble with this insect stuff is that you can do crap work and get published. If you do work on robots that really balance, you look stupid if your control system doesn't work. Everyone can see you failed. With insect robots, failure is less obvious. Some people think this is a feature.

Biggest application: NASA

[mblase]

There's way too much work on insect-level locomotion. Brooks was doing this stuff twenty years ago, and took it about as far as it's going to go.

I don't think that's true. There was an article in Discover a month or two ago (can't find it online, sorry, but I believe it parallels the linked article) where a researcher was trying to tease more information out of a cockroach's walk, discovering that it doesn't actually use a three-feet-down-all-the-time approach but wobbles side to side, remaining dynamically stable as it walks. This is not what you might intuit by simply watching insects walk.

As for "too much" being done, I must disagree. Walking robots aren't as good as they can be or it'd be perfected by now. Wheels are faster, but only over ideal terrain; complicated terrain that would confound the best wheels can often be navigated by legged animals. NASA's interplanetary rovers all use wheels, and all of them eventually encounter situations where they're useless, so if they could deploy a robot lander that could walk effectively (and efficiently), it'd be of tremendous value to them.