DARPA Puts $32M Toward Quadruped Robot Prototype

The Installer writes with this selection from GizMag: "Walking quadrupeds are being cast to play a major role in the rapidly unfolding age of robotics. The platform promises versatility far beyond that of wheeled-vehicles and will undoubtedly find applications in a wide variety of fields. Not surprisingly, the development of quadrupeds is being driven by the military and DARPA has recently boosted its efforts by awarding Boston Dynamics $32 million for the prototype phase of its Legged Squad Support System (LS3) program. ... LC3 is conceived as an autonomous support pack-robot for ground troops that can carry 400 pounds or more of payload, sustain itself for 24 hours and cover 20 miles in almost any kind of terrain."

Comments

Why four legs?

[immaterial]

Wouldn't six be more stable, distribute weight better, and be more redundant in case of failure of a limb? The only advantages I can think of to four is better efficiency (maybe? Having to move fewer limbs should be a plus, but on the other hand they'll each also have to be more powerful) and a slightly more compact form. Is it worth it? I'd certainly find a couple extra limbs convenient; you'd think a military robot in potentially harsh conditions would also find a couple spares very useful.

Re:Why four legs?

[immaterial]

Well, was there a time when vertebrates ever had six limbs? I thought we'd just happened to pop out of the water that way; given that we vertebrates have all kept our basic layout since then, I wouldn't really say evolution ever had the opportunity to try six limbs on a large land creature.

Obviously more legs doesn't necessarily mean better. Why not eight, like the last guy said? There's an advantage to redundancy but like you said there are disadvantages as well. It just seems to me like six would be a happier medium between the two.

You need three points on the ground to remain relatively stable without getting into balance issues. Of course a good robot could balance itself on two - hell even a Segway can do it - but lets say in a complex situation like trying to climb a steep rocky hill in Afghanistan in a hurry you want as much stability as possible (note a person in that situation will be using their hands for additional stability too). To move while doing that requires at least one more limb, for a total of four. If there is have a failure in just one limb, you've reduced the robot's stability by quite a bit. I just seems to me like six would be worth it in varied terrain - I mean, if all you're going to do is bound across the plains all day like a gazelle the extra limbs may not be useful, but I'm imagining these things in more steep, mountainous areas as well, where extra footing could be a huge asset.

Re:Why four legs?

[Animats]

No, four legs are better for a large machine. There's a tradeoff between leg working envelope, vehicle length, and top speed.

There was a big fad for six-legged insect robots in the 1990s, led by Rod Brooks at MIT. Those were very slow, very dumb, and had a very wide stance. Six legs don't scale up well. One big issue is inertia.

Double the dimensions of something, and it gets four times as strong (strength comes from cross-section) but eight times as massive (mass comes from volume.) This is called the cube-square law, and it's why there are no giant insects. For small creatures, forces like surface tension matter, but inertia doesn't. For large, fast ones, inertia dominates.

Before dynamic balance was figured out, robots tended to have very wide stances, and some had too many legs. DARPA built funded the Adaptive Suspension Vehicle at Ohio State in the 1980s. 28 feet long, six legs, seats one, no cargo capacity. Top speed 3-5 MPH on flat ground. At least three legs were on the ground at all times, and often four, five, or six. The gaits were very conservative. It was supposed to be off-road capable, but that part never worked. A sloping road was as far as they got. There was some computer control, but the thing was mostly driven by an onboard driver, using three joysticks.

With dynamic balance and traction control, the leg geometry doesn't have to be as conservative. BigDog's leg geometry is four legs with three joints each, a narrow stance, and control which allows the leg envelopes to overlap. This is close to the layout of the larger quadrupeds. (BigDog has the size and weight of a medium pony; it's bigger than dog-size.)

With four legs and a long body, pitch stability isn't too hard, but roll stability requires active control. The faster quadrupedal mammals have very narrow stances; a horse's track is less than a foot wide, narrower than its body. BigDog doesn't track quite that narrow, but it gets close. The narrow track makes tight turns possible, and allows sudden changes in yaw when needed for slip recovery or collision avoidance.

With dynamic balance and slip control, the speed can be cranked up. The six-legged machines mostly crawled; the modern four-legged machines trot, and some run. (The usual running gaits, the ones with a moment of suspension, for a quadruped are the trot, pronk, rotatory gallop, and canter. BigDog can trot and pronk; it may be able to do a rotatory gallop.) That's the real reason to go with four legs. Six legs just get in the way at speed.

BigDog's three-joint leg isn't mentioned much, but the third joint lets the control system adjust the ground contact force vector to stay within the friction cone, without changing the foot position. This is a big win when climbing hills, and the hind end needs to come under the body.

It's all about the control algorithms. Don't let the legs collide, prevent slip, recover from slip, support the body, maintain roll balance, provide propulsion, avoid obstacles, stay on course, accomplish the mission. Those are the priorities.

If you want to understand the theory behind BigDog, read Didier Papadoupolis's thesis, "Stable Running for a Quadruped Robot with Compliant Legs". The technology for BigDog came from Martin Buehler's lab at McGill University. Buehler himself quit McGill and went to work for Boston Dynamics as the chief engineer on BigDog. (Once BigDog worked, he went to iRobot.) The theory is out there in the literature. Some of it is mine.

Duplicate story

[ajknott]

Already been covered - http://tech.slashdot.org/story/10/02/01/2141213/Militarys-Robotic-Pack-Mule-Gets-32M-Boost

Congratulations

[srussia]

from TFS: LC3 is conceived as an autonomous support pack-robot for ground troops that can carry 400 pounds or more of payload, sustain itself for 24 hours and cover 20 miles in almost any kind of terrain

They just specced a camel!