MIT's Cheetah Robot Runs Untethered

An anonymous reader writes: It's easy to make a robot walk, but hard to keep it from falling over. We've seen a number of crazy robot prototypes, but they're usually tethered and/or stuck on a treadmill. Now, researchers from MIT have developed an algorithm that allows their giant robot cheetah to run around outdoors at up to 10mph. They expect the robot to eventually hit speeds of 30mph. "The key to the bounding algorithm is in programming each of the robot's legs to exert a certain amount of force in the split second during which it hits the ground, in order to maintain a given speed: In general, the faster the desired speed, the more force must be applied to propel the robot forward. ... Kim says that by adapting a force-based approach, the cheetah-bot is able to handle rougher terrain, such as bounding across a grassy field." The MIT cheetah-bot also runs on a custom electric motor, which makes it significantly quieter than gas-powered robots. "Our robot can be silent and as efficient as animals. The only things you hear are the feet hitting the ground."

Comments

Of course you use force control to run fast.

[Animats]

That article is written as if that crowd invented running using force control. Of course you use force control. Everybody in the field knows that by now. I patented that 20 years ago. The Scout II robot at McGill, developed by Prof. Martin Buehler, used that approach. Buehler went on to become the designer of BigDog, but never got much public credit for it and quit to work for iRobot.

The key to legged running in non-trivial situations is careful management of ground traction. Traction is first priority, then balance, then foot placement. Historically, everybody worried about foot placement first, but that turns out to be backwards. As soon as you get off flat surfaces with good traction, traction control dominates.

The next unsolved problem in that area is not going fast. It's starting, stopping, and turning fast. Most of the legged robots accelerate very slowly, and don't make abrupt high-speed turns. Big Dog starts by trotting in place, then extending the gait out. Starting fast, stopping fast, and turning fast are all facets of the same problem. You have to take one stride using completely different control algorithms than you use for normal locomotion. That's all I'm going to say about this for now.

Re:Of course you use force control to run fast.

[Bill_the_Engineer]

MIT news always gives the impression they invented something new. I remember a couple of years ago when NPR did a story about some MIT students launching a weather balloon with a camera attached and took pictures from the edge of space. They even went as far as suggesting that one day NASA could use this technology to perform experiments cheaper.

I was taken aback by this announcement, since:

1) Who hasn't launched a weather balloon with a camera attached?

2) (The thing that really made me laugh) I was waiting for the weather to become favorable for our research gondola to be launched at NASA's balloon facility.

Re:Of course you use force control to run fast.

[Animats]

Pardon my ignorant question, but how is it a problem to have traction control? Wouldn't it be enough to glue traction strips to the feet or something?

That's like wearing shoes with golf spikes all the time.

Traction control for feet does roughly the same thing as automotive traction control for cars. The basic idea is to keep the sideways force below the break-loose point. This is the down force on the wheel times the coefficient of friction.

For car wheels, the down force is mostly constant. For a legged robot, it changes throughout the ground contact phase So the side force has to be actively controlled and changed throughout the ground contact. It's also necessary to compensate for leg angle.

Legs have an additional option. If a leg has three joints, you can adjust the angle at which the contact force is applied. This is a big win on hills.

I used to work on this stuff in the mid-1990s, but nobody was interested in building legged robots back then. It could be used for animation, but it was overkill for games. I never expected that DARPA would spend $120 million on BigDog. Robotics projects in the 1990s were tiny.

Excellent

[mentil]

Now all they need are robotic gazelles. Its silent electric motor lets it sneak up on its prey on the robo-savannah, consuming its delicious battery juices so it can live another day.

Re:Asian-only team?

[vivian]

One of the things that makes US reasearch strong is the ability of it's universities to attract te best and brightest from all around the world. This is nothing new - it has always been thus - though perhals this is incresingly so as the state of secondary education seems to be in decline compared to opter parts of the world.

When researchers stop coming to the US, the state of
research there will go into rapid decline.The US isn't alone though - it's the same story in Australia too.

Given the already established centres for excellence in the US, it's a favoured destination for smart and motivated people from India and Asia, as well as other parts of the world to further their education and opportunities.

Half the world's population is Asian or India/Bangadesh/Pakistan, so naturally you are going to see many from those regions. Be glad for it - or they would be busy innovationg wherever they came from instead of the US.

Re:Asian-only team?

[rioki]

European researchers work at MIT and nobody beats an eye,
Asian researchers work at MIT and everybody looses their mind.

Re:Implementation Fail

[wonkey_monkey]

If this is an implementation of a cheetah's running locomotion

It's not...

it is a complete fail.

...so it's not.

Re:Implementation Fail

[Sockatume]

That's because the robot's not galloping, it's bounding. It's a completely different gait.

Believe it or not, people who study animal locomotion do actually do research on animal locomotion.

Re:Implementation Fail

[Rashdot]

So they should rename it to cheatah?

Re:Asian-only team?

[qbast]

Just wait until African researcher works at MIT ...

Not a cheetah

[Cytotoxic]

Exceptionally cool and really fast legged robot. I really liked the ability to alter its gait to bound over an obstacle.

But the defining characteristic of cheetah locomotion is not just the 70mph speed. It is the use of a long, flexible, back to power and lengthen the stride. This bot has a completely inflexible body and is solely powered via the leg joints. Not very cheetah-like. More like a sheep.

But sheep-bot is just not very cool. And if I had built the thing, I would have named it a cheetah robot too. Or maybe Mechanical Hound or Robot 451.

Re:Asian-only team?

[AthanasiusKircher]

European researchers work at MIT and nobody beats an eye,
Asian researchers work at MIT and everybody looses their mind.

Uh, who is "everybody"? Certainly not MIT itself. Stats on their international students show that about 50% of all international students and scholars come from Asia, much more than Europe.

MIT cares about "diversity" numbers, sure, but they already can claim that they are a "minority majority" campus with over 51% of undergraduates from minorities. So, they'd really have no reason to further inflate the Asian numbers... unless, well, the Asian students were actually more qualified.

Which means Asian researchers are probably working at MIT because MIT actually is looking for highly qualified people -- and thus, MIT hired/admitted them.

The only people "losing their mind" are racists, who clearly aren't "everybody." (If they were, MIT wouldn't have such high numbers of minorities in the first place.)

And why are we discussing this anyway? TFA has a photo of the lab team, which is certainly not all Asian in composition.

oblig.

[seven of five]

http://www.youtube.com/watch?v...

Been there done that.

[digitalPhant0m]

Hasn't this already been accomplished a few years ago. Boston Dynamics has been doing this for awhile.
https://www.youtube.com/watch?...