this thing is in the news again, but I first saw pictures of it three years ago.
The researchers say it could take another five years before the robot can be used by the elderly. At present, the device is slow and needs modifications to carry heavy loads.
I think it's time give up, guys. Or at least stop trotting out the same demo over and over. Or at least dress it up.
Why not just get an Ibot?
Hah, I didn't even notice that it was in your three years ago entry... When living in Manchester (where DEKA is located) I see these all the time, along with Segways, being tested.
Saw an Ibot freak out while they were testing new code.. very funny.
Whoa, at first I thought that said 'Emotionally friendly hand grenades...'
And copper isn't cheap.
Can we used recycled plastic in those grenades? Why not just practice with wadded up plastic bags, collected from greenways around large cities.... First, recruits spend hours collecting the materials, then they throw garbage at each other....
Nevermind.
(Presumably, banned after this inanity.)
So pretty much everywhere I go in Japan is low ceiling and has almost absurdly small doorways, someone cruising around in one of those is never going to able to go inside anywhere.
For reference I'm a middling 5'8" and I have to duck all the time.
Is that "mechagimp" or "megachimp"?
These guys are never going to get there. We don't even understand how humans walk from a biomechanics standpoint; we have even less idea how our brains go about controlling the whole process.
These guys will get beaten to it by the exoskeleton people, if only because they've skipped the control step entirely. These guys still have only had marginal success and with unpowered assist structures.
These guys have failed at everything except getting money. The things that will make a biomimetic system work have to do with understanding or at very least matching the performance of the component systems: muscle, brain, and metabolism. Hydraulics, electric motors, computers, engines and batteries are such amazingly poor substitutes that you'll never build an android with them. Fuel cells may — and I do mean may — give you sufficient power density. The other two? Nothing in the pipe. The last "artificial muscle" paper I saw came dangerously close to breaking the Second Law of Thermodynamics.
Hydraulics, electric motors, computers, engines and batteries are such amazingly poor substitutes that you'll never build an android with them.
Electric motors are way heavy, I agree. But I hope you're proven wrong about hydraulics in the next couple of years. The historical problem with them is not lack of power, but rather crap control systems. Good hydraulic servovalves (check out Yobotics) are capable of loop bandwidths in excess of 30 Hz, which is plenty for the majority of human movement.
As always, the stumbling blocks are power plant and AI. I don't think we're going to solve the AI problems in my lifetime. But provided you don't care about efficiency, the power plant problems are mostly a matter of aesthetics.
Boston Dynamics already has an autonomous four-legged robot that trots stably under its own power, and has excellent balance. It uses a model aircraft engine. I found some two-stroke 2 cyl model engines that weigh 9 pounds and put out 8 HP. They're noisy, sure, but they will power your development platforms and are more than light enough. And up the scale a little bit, small turbines exist that have power to weight ratios of nearly 2:1. A Wren MW-54 free-shaft turbine weighs 5 lbs and produces 8 HP. Again, noisy as hell. But for prototypes? More than good enough.
Totally agree with you when it comes to the exoskeletons, though. Sarcos Corp's exoskel is already supposed to allow people to carry 90 kg loads as if they weigh 5 lbs.
http://www.ai.mit.edu/projects/leglab/mpeg_vcd/videos/hea.mpg
These are the muscle analogs that will allow robots to be both dextrous and powerful. They allow you to deliver a pound or a hundred pounds of force with no humanly observable difference in reaction time. They can apply constant force even under changing loads and degrees of extension.
Their one downside is that you're not guaranteed that they'll be exactly at the spot you want them to be. But that's true with human beings too - we have a visual feedback system that we lean on heavily. So the big missing piece right now is actually artificial visions systems. All the robots we build are still blind, and hence clumsy. They're groping around in the dark, with no idea what they're reaching for. Robots will continue to be suitable for sandboxes only until we break that obstacle.
I find it hard to believe that the level of "AI" necessary for working a leg are really that far off; that strikes me as the sort of problem that, worst case, you can just throw cycles at, so Moore's Law. And better-case, it smells like just a matter of finding a shortcut out of the combinatorics. I mean, this isn't consciousness, this is all autonomic.
Granted, I don't know anything about the mechanical engineering side of this, but I think I know enough about AI to know that "AI" means little more then "software we haven't written yet".
The power side does sound like a bitch, since we've got nothing near the efficiency of biology yet. But if the last problem remaining in this is battery weight, that's still awfully far along...
Also: I can't find it now (I wish lj had a way to show me tag intersections, e.g. "robots" and "mpegs") but a while back I posted a video of this robot that managed a very human-like gate unpowered, so it's possible that a big chunk of this problem is really just weight distribution and static balance: solve the math once and replicate.
Motion planning is a PSPACE-complete problem, and thus difficult to solve no matter how many cycles you throw at it.
And yet lizards manage to do it, and they barely have two neurons to rub together...
Lizards don't have that many degrees of freedom to worry about, days to teach their neural networks, and billions of years of evolution.
They also don't worry about satisfying all conditions before moving. Biological systems tend to discard huge amounts of information - either conciously or at a precognitive level. Which is good, if most creatures are presented with too many options, especially options which aren't sufficiently distinct, they tend to be unable to make decisions effectively (which is why you never want to give someone too many options because they'll likely choose none of them).
Right, whereas if you get even one robot kicking a child to death, people get unaccountably annoyed.
Your point about not needing a lot of AI to do basic walking is well taken. I think a big part of the problem that robot builders have had with natural walking has been, ironically, their attempts to exert too much control over the joints and legs. Trying to "muscle" the robot into a certain kind of walking, with a high degree of precision in where each foot goes and how much the joint bends and so on. Which inevitably fails if the floor is a little bit unlevel, or a motor doesn't respond quite as fast as it used to, or etc.
"Passive dynamics" is the new buzzword here. The idea that you figure out how your robot would naturally walk, and arrange the muscles so that they help it fall into that kind of dynamic state.
The robot you were thinking of might have been Andy Ruina's passive dynamic walker:
http://www.sciencenews.org/articles/20050806/bob8.asp
Though a quick google found at least three more passive dynamic walkers, I also googled "jwz boot face site:livejournal.com" and found this entry. Haha.
Anyway, when I said the AI problems won't be solved in my lifetime, what I was trying to say was that I'm betting against living to see human-level intelligence in robots. Walking, that I think we can figure out. I mean, if roaches can do it...
Yeah, that's the entry I was looking for...
Ray Kurzweil is predicting the singularity in ~2044. I'll see you in the Matrix!
The best thing about the 'green' grenades?
The plastic will likely be x-ray transparent so you get the killing power plus wounds that are more difficult to treat through conventional means. Good for the environment and increases the logistical impact of a grenade.
Its a win win situation!
I don't get this: what kind of a plastic has a density anywhere near that of copper? No density, no momentum, no penetration.
Well, in war time the real trick isn't to kill your opponent but to wound them. This impacts the war fighting capability because it takes a soldier out of the lines for a certain period of time, ties up an personelle used to bring them out of the front line, uses up hospital resources, and ends up being demoralizing and long term drain on the economy. Don't get me wrong - killing them is a good time but overwhelming their medical support system is priceless.
As to the specifics - there isn't any plastic with the same penetrative capabilities as copper but you can do more with plastic. You can get a reasonably dense plastic and then construct it in such a way that the grenade casing basically shatters into thousands of tiny needles moves at several thousand feet per second. It would likely effectively flense any exposed skin or lightly clothed extremities. Of course, this is all in addition to the normal blast effect (which can be augmented by increasing the amount of charge since the grenade is now lighter).
This is all over the news again here in Japan. I wonder why they are enjoying a second, late, mad rush of publicity.