I thought Toyota was pretty good until I followed the link in the article to Big Dog. http://www.youtube.com/watch?v=W1czBcnX1Ww
Big Dog is truly awesome. It runs, jumps, recovers from slips, climbs over rubble, ...
I ran into an MIT engineering grad who was talking trash about Big Dog. He said "When you see a video of a new robot, remember that what you see is all it can do." In the case of Big Dog, this apparently means that it can run but not walk. And we only get to see the runs where it doesn't collapse. OTOH, it is truly impressive and sinister.
Amazing what an Industrial Military Complex can do with half the US Budget. I'd like to see them put some money into space myself but robots are good too.
Arguably, putting money into robots is putting money into space. Along with a lot of other things, too. Robots have already long since passed through the hype cycle and now we're firmly sitting at the end of the cycle in "ho-hum", but it's worth taking a moment to realize just what this all means, and that the old sci-fi predictions may not have happened on schedule but are about to start popping up.
While everyone's busy moaning about the economy and how terrible everything is, I've been looking down the road five or ten years and I see amazing things coming at us, fast.
I don't see a lot of household robotics in that time frame, but I think we're going to see a new generation of industrial robots. Some of them have already started to pop up, and robots like this story will continue that trend.
Right now, robots only work in spaces explicitly designed for them, which has really limited their effectiveness. They're going to be able to move beyond that into more human spaces, which has two major advantages: A lot of human spaces exist, and those spaces are set up to do more different things, which robots will be able to participate in. An assembly line is a fine thing, but not terribly flexible. (Though increasingly, this is a fuzzy line, and the line will get fuzzier; indeed, this is one of the effects I am anticipating.)
Computer vision seems to have taken a new step forward lately, and while they're still miles away from "full human vision" (which is AI-strong, ultimately), they're moving from total toy problems into really useful problems.
Brain interaction is also moving forward lately.
Now, you might say I'm not being specific. Well, here's why. In the five-to-ten year timescale, what I see is that where today these things are still the cutting edge of research, in the five-to-ten year timeframe this is going to be available to small startups and dedicated hobbyists. I don't know exactly what will be made of advanced computer vision and smart robots and direct brain interaction and another five-to-ten years of hardware development and two guys in a garage (oh, and we'll still have the internet, of course, and pervasive cellular connections), but I find it quite likely that it'll be very significant... in exactly the same way the Internet is significant. It may take a genius to push the frontier of computer vision or robotics, but thanks to the miracle of programming, it won't take geniuses to apply them in new ways.
Also, I could probably go on for another three or four times longer, but this is an HN comment, not an essay. (Though maybe I should make it a blog post.)
You wake up today and it's much the same as yesterday, but you know, look back at what the Internet has done to so many industries in the past ten years, and what it has yet to do, and I think you start to get a reference frame for what we may be looking at with the convergence of some of this stuff. I know I'm not giving you product names and the bullet-point feature lists, but that would be almost exactly like me trying to predict Facebook back in 1990. Even those who were on the right track in the broad sense were wrong in many ways... but they were right that big things were coming.
Here's a specific technology that's going to be a huge boon to robotics in the next 5-10 years: Flash LIDAR cameras. This is the technology behind the XBox 360's "Project Natal", but it's a lot more than a game controller. A Flash LIDAR camera directly senses the distance to every object seen by the camera by measuring the time-of-flight of a reflected laser pulse. With accurate distance data for every pixel a lot of really hard computer vision problems suddenly become much, much eaiser.
One thing that might happen within the next 5 years is that robots start to learn more from what they see in their environment and from the feedback they get. I'm not sure how much that is already part of current robots, but I haven't heard about it so I think it's not an important part.
One way we humans learn is by observing things, trying to immitate them and learning from the feedback (either physical or via communication). If that principle takes hold in robot development, the "evolution" of robots could get much speedier I believe.
The ability to act in the physical world and actually cause things to happen, cause reactions, is what differentiates robots from other types of computers. So that's what I would try to exploit thinking about the principles of programming robots.
I wouldn't be surprised if it uses neither. My guess would be it has encoders to know the state of the joint angles (so, a type of proprioception) but probably not for foot contact.
If someone has a reference to some technical details, though, I'd be very interested.
Thanks for the link! I just figured it would use the simplest solution that could get the job done. I don't know how important measuring foot slip/ground contact is for quadrupedal locomotion or how well it can be approximated by measuring joint angles and computing forward kinematics. I guess load cells are cheap so why not use them?
Again, thanks for digging that up. It's heaps better than idle speculation.
Getting the job done is important! A simple solution is no good if it doesn't work. Presumably they need to know where the ground is / how firm the ground is, before putting all it's weight down. There is a good bit in the video of it climbing over rubble.
I am not interested in a roomba because I don't see it getting in around and behind the toilet, the place I most hate to clean. Sweeping the wide open floors is easy enough to not even be worth $50.
But something like the above, that has:
Enough hand eye coordination to catch a rolling ball.
Enough facial recognition to sort of be able to tell people apart, like a human who's bad with faces.
Enough speech parsing to understand simple command even if you have to repeat them a few times.
Enough knowledge, I'm thinking something like Wolfram Alpha here, to able to be instructed to perform simple tasks.
Then it could do simple tasks like clean behind the toilet or work the fryolator at McDonald's. And if it's mass manufactured and cheap enough to build and cheap enough to buy, then we could see something interesting.
The slow uptake of humanoid robots into the everyday workforce. How long before we have robot janitors?
It single-handedly licked my kitty-litter tracking problem. it keeps my place completely swept on a weekly basis (I've a small studio loft, nothing huge) - preventing me from stepping on the odd tiny rock that was drug in by my workboots. It devours dust/hair bunnies which come with long hair and a cat. I never swept so the 120$ woot price on the roomba has changed my place from gritty to well kept.
Altho damn, now I wish it scrubbed the floor behind the toilet.
Back to the meat there - I think that the next advance in robotics is going to be alongside artificial limbs and human assistance devices. AI introduction into society seems to receive such strong philosophical opposition that I doubt we'll see much in the next 10 years.
The high-speed footage was most interesting to me. In terms of the "uncanny valley" and my personal reaction to the robot's movement, I find that it's still not human to me... but I think it's starting to crawl up the side towards human-looking. In terms of the traditional graphic (http://en.wikipedia.org/wiki/File:Mori_Uncanny_Valley.svg ), I think this thing is somewhere between the prosthetic hand and the bunraku puppet, on the higher end of that range.
I wonder why they don't build humanoid robots more like monkeys. The reason for robots to be humanoid is to enable them to move in spaces made for humans, right? Monkeys can do that too, only better, because they have much longer arms and can use their arms to help with walking without falling as well as reach higher up without being taller.
Sometimes I get the feeling that making robots as similar as possible to humans is a goal in itself, and that, I think, is not a useful dogma.
Hm, if monkeys are better suited for human environments than humans, why don't we look like monkeys? There must be some advantage to human proportions?
Which one? You're implying that all things are best the way they are because everything trends toward perfection. Having tried quite a lot of office furniture recently I can report that humans have not figured out what's best for human proportions :)
The Asimov stories illustrate exploits on his own 3 laws.
But how can you fix that? Embedding a full robotic-oriented legal system in their little brains does not seem to be the answer neither.
I think we have to admit that robots will be far from perfect. They will make mistakes. They will stumble and fall. They mill misjudge situations. Like we humans do.
Not to mention that creating a brain with the three laws ingrained would be an order of magnitude more difficult than creating a brain in the first place.
Interesting that it runs on the balls of its feet rather than using the heel-first running that running shoes tend to promote. Is the barefoot ball-first gait easier to implement?
The ball-first movement is much more natural for humans too! Running shoes promote rather un-biomechanically sound forms of locomotion. I think in 100 years modern shoes will be considered with the sense of horror footbinding is regarded today.
[Eric Cartman voice]
"You can trust Awesom-o. In fact, you should tell Awesom-o all your most personal secrets. Awesom-o will not make fun of you or tell your secrets to other people and stuff."
