(RANT) Androids: Why don't we have some already?

[Nyx]

I found out about this site a few days ago, and it somewhat revived my interest for robotics. Ever since I was a kid I always found mechanical, electronic, logical and scientific things interesting. Robots were a bit of all of this. Unfortunately, I never really had the chance to go and build a robot, mostly for financial reasons... And still today I am debating whether it would be a worthwile investment to spend at least $300 building a mobile "drone" robot, which would still have to remain rather simple, as that would only afford me a mobile platform, two color cameras, and an interface for my laptop to connect to it and serve as the "brain" of the robot.

It seems the development of innovative robotics is mostly left to the big players. We will occasionally see a flash on TV about Honda's ASIMO robot... Which cost so many millions to develop by a huge team of engineers, and yet seems to be surprisingly inneficient and primitive. Everything ASIMO does has been done years ago by researchers in universities around the world... It is by no means revolutionary, and only seems to serve as a publicity tool for Honda, rather than a serious research platform.

We also see a few robot expos here and there, but what they feature seems to be mostly automatons, and not actual robots. Or if they have robots, they are usually clumsy, bulky and impractical, with an AI so limited it would have zero use in the real world. They don't really have much "intelligence", and their level of mobility is laughable. Kind of saddening... But, is this really the state of the art?

---

Honda qualifies ASIMO as an "android". But seriously, where are the androids from the movies? The ones that look human, have some degree of logical understanding and intelligence, can speak and understand what people say, and are autonomous? They're really nowhere to be seen. There are lots of projects floating around, but it seems nobody has actually ever completed one.

Is it because of funding? Probably not. Honda seems to pour quite a bit of funding in the ASIMO project. And I'm sure alot of companies out there would love to have their name associated with the first viable android out there.

Is it because of a lack of technology? We have powerful energy-saving dual-core processors, very good voice synthesis (ATT), decent voice recognition, relatively good binocular vision, moderate AI, compact electric muscle fibers (muscle wire), light and resistant materials (titanium, fiberglass), the capability of creating human-looking "shells" (I would name a reference here, but I better not)... We have decent power technology (compact and light li-ion batteries), we have high-resolution, high speed color CMOS cameras, tiny pressure sensors... Self-balancing low-energy walking machines (see passive walking)... And more... We really have everything it takes!

Of course, the complexity of such a project is enormous, but don't tell me it can't be done at this stage... Don't tell me it's because of power consumption... Because there are tiny laptops with huge processing power that can last 3 hours on a tiny battery... And walking requires a very low amount of energy if done right (eg: self-balancing, instead carrying load like industrial machinery).

Fitting all the required components in a human-sized frame should also not be that challenging. What would take the most space is the batteries, followed by the onboard computer, and the huge number of electrical wires required. And if electrical muscle fibers were used, they wouldn't take much space or require a pressure pump like hydraulic actuators. They would only require some moderately complex control circuitry, which can probably be fitted in a very small casing using modern electronics.

What is challenging is the engineering and programming. The actual electronic components for such a robot would actually be pretty cheap (most likely under $100K for it all), but the engineering costs would get quite high... But like I said... How many millions has Honda shoved in the ASIMO project thus far?

---

The way I see it... There are a few very simple reasons why we don't have advanced androids... And they are not due to technological or financial limitations.

1) Our society has no absolute need for them. We find them fancy and beyond cool, but do we need them? This will not sound politically correct at all, and I do not agree with that view, but our society feeds on cheap labor, and to capitalists, it seems the life of an african or mexican is just so much cheaper than the creation of an android. We could use androids for hard, risky and unpleasant labor, sure, but paying someone from a foreign country to do it will cost you 50 times less. Nobody would say this out loud in a public speech, but it's clearly reflected in the actions of large corporations, and our society in general. The only "need" for androids is the one we artificially create, but since our world feels it is cheaper to get by without androids, it gets by without them.

2) Scientists, engineers and programmers are lazy and scared. That's right, they're pussies! The creation of an android is a project that promises great perspective, and those who will (eventually) succeed at this will make it into the history books.... BUT... The project also entails huge *risk*. The risk of failure is high, finding funding and keeping it is hard. Creating an android is a life project. To do it, several researchers, engineers and programmers would need to dedicate a large fraction of their life to the project... And if it were to fail, they would risk losing everything, they would risk being regarded by others as failures who have wasted time and accomplished nothing. Furthermore, the complexity of the problem at hand is great, and most people in the area have difficulty foreseeing how it all fits together. This scares alot of engineers and programmers because it is far from the relatively trivial tasks they are used to (which they almost always succeed at).

3) Our capitalist society has a problem with the notion of "long term benefit". While it's true that androids could benefit us in the long run, people have an issue with that. Investors don't like it, because if they invested in an android project, not only would there be a huge amount of risk, but it's possible they could die before the benefits of such a project concretize. To an investor, this is just not worth it. Anything that can't translate in terms of investing some X amount of money now and obtaining YX back (where Y > 1) within the next 5 years is simply not worth the risk and the wait. They would rather invest their money where they are sure that they can obtain benefits quickly. Whether what they invest in could benefit mankind or help science progress further is irrelevant to 99.9999% of them.

4) There has been a crash in the level of interest of the public. This is often disregarded... But it's true. During the 20th century, technology progressed like never before, and this raised everyone's hope for a much better future... A world where science would bring us miracles... We saw science fiction movies and books that depicted such worlds.... We saw all that 50 years ago that is. Science fiction came very rapidly. Asimov's wrote sci-fi novels between 1939 and 1958! He was already able to foresee the possibilities robotics might bring us before the field of robotics as we know it even existed. And then there was Star Wars, and many others. All depicting robots with a level of advancement beyond anything that was to be seen in the real world. I believe people got "bored". The idea of advanced robots and androids sparked an interest due to their novelty, but since people haven't seen anything close to that in the real world, the public interest shifted to other areas. The robot sci-fi "fad" lost some of its importance over time. Up to now, where people don't talk and think of robots that often anymore. People, I believe, consciously or not, see androids as a "failed promess". People have been telling them "it's coming quickly and it's going to change our lives forever" for the past 70 years, but this just didn't happen (yet)... Science "failed" to deliver it on time (just like the flying car).

I still believe there is hope, however. Research is still being done in the field or robotics in every big university out there, and this means the technology can only get more sophisticated. One day, the components required to build an android might become so cheap, that someone may be able to build one during his space time in his garage... And in that sense, it's bound to happen someday, even if that day may be relatively distant... And who knows, maybe one of these android projects will be able to get their act together and succeed.

Discuss

[Admin]

wow ok you said a mouthful . . .

about funding robots, you might want to read this:
http://www.societyofrobots.com/robotfunding.shtml
I would say all the time/money I spent on robots when I was a poor college student was the best financial investment I ever made. The experience I gained has given me my nice pay checks in the proffesional world 

What makes building androids (robots that act like humans) so hard is that humans are extremely robust. We arent good at any one thing extremely well, but we are still good at a huge number of abilities (running, memory, vision, climbing, swimming, planning for the future, regeneration, etc etc etc).

But we have built machines that can out perform humans in each of those . . . cars are faster than running, computers remember way more than people, we have night vision and infrared cameras, boats are faster than swimming, the list goes on. Robots have already taken over the factories of cheap labor.

What use is an android when we have all of that?

But we also cant build something human like because we ourselves still havnt figured out how we humans work. How the brain works is still a mystery to our scientists, for example. We also havnt figured out how to make artificial muscles that have comparable abilities to living tissue. And we definitely havnt figured out how to make robots self assemble or regenerate . . .

As for your other points:
1) Capitalism isnt as bad as you say. If I had a choice of hiring one of two robotics engineers, one an american demanding $50k/year and the other (a mexican, lets say) $30k/year. Lets say they have equal skills. Heck, why would anyone hire the american? This doesnt just work for 'jobs no one else wants to do.' It works for high skill labor too . . . If the robot building robot costs $20k/year, then I just wouldnt hire any humans anymore . . . Unless of course its like bomb disposal or nuclear cleanup that just kills off my employees . . . then Id get the more expensive bot . . . dead employees = bad

2) There are a huge number of android projects out there (like 50 just in japan alone). Dont think they are afraid . . . but businesses that pay for these projects however are afraid that they cannot profit from the expense of the research. Honda and Sony have spent a decade or more on this, still no profit.

3) Yea I agree. Ive heard most businesses dont look more than 5 years in the future. The problem is paying costs for 5+ years without making any profit, its really hard. What happens when your 7 year robot project has one year left to go, but you run out of money to pay your employees? They quit and your business is screwed . . .

4) The problem is uninformed futurists and science fiction writers . . . I personally claim we have at least another 50 years before we have androids running around . . .


In the end, its economics. If me buying a robot can save me more time and money then it costs to pay for it, then Ill buy it. I am sure you agree . . .

[Arislan]

This seems to be about more than androids, there seems to be something about human-like AI in there...but anyway...

Nobody is stopping you from starting an android project. People will say it's hard, and it is, but no one will flat out tell you not to try it. So go make one.

[Gopher]

Just wanted to add to Admin's comment on #3. Japanese businesses plan much further ahead than US companies, some as far as 50 years ahead. This is why Honda and Sony are willing to spend so much money on these projects for so long, they envision it giving them an overwhelming advantage when the market does mature. Honda actually started the Asimo project 20 years ago; the closest to that commitment I've ever seen from an American company in modern times is the guy who's been selling shares to investors in the same flying car since the 70s, but most engineers dismiss him as a con artist. I'm not sure if it's good for business or not, but it's definately good for technology.

[Arislan]

I remember a flying car in the 70s that looked like a rounded platform, operated by a guy standing up.

[Admin]

Ehhh but Sony and Honda have really deep pockets, they dont count. 

Smaller companies simply cant afford to invest long long term, even if its wiser - you have to pay the bills in the short term too . . .

Google thinks long term, no?

[Gopher]

Arislan: Same guy is still around. That incarnation was called the Discojet, the new version is the Skycar. New look, new marketing angle, exact same claims; he's steadily bumped back the milestones every few years, the only thing new is an integration of information about new FAA regulations regarding new smaller classes of personal aircraft and a lot of talk about complete autopilot systems. Check it out on Wikipedia, it's amazing how successful he's been at collecting investments over the years despite making no demonstrable progress and many demonstrably fraudulant claims.

Admin: I may be wrong, but I believe I read an article (I forget what magazine but it might have been NewType?) which claimed that companies in Japan typically planned their business stragety on a substantially longer term than their western competition. I was only citing Sony and Honda as specific examples related to robotics, though you are right that the smaller a company is, the more of it's resources must be committed to the near-future. Of course, I may be misremembering, or the author may have whipped it out of their arse. It was also a number of years ago, and the recent economic pressures have shifted priorities for everyone; I was sad at the discontinuaton of Aibo, it had just started to get really interesting in the last couple of generations.

[Arislan]

Hi Gopher,

No, the one I saw in the 1970s was the "Hiller Flying Platform". As for the Discojet and Skycar, don't they work? There are pictures of them hovering online...faked?

I think Honda is right in planning for the future, I just hope we are all alive when that future comes. I'll happily have an android that looks like actress J.R. Carrington

Lest I forget, Sharper Image and a few other stores sell a robotic chimpanzee head. Pretty interesting.

[Gopher]

No, the hovering is real, but for a quarter of a century they've claimed to be on the verge of getting it to transition between hovering and really flying. In the hover mode it is very slow and impractical, a novelty rather than a form of practical transportation.

I believe the Hiller platform was even older,  dating all the way back to the 50s? But it's basically the same thing, without the posturing and slick marketing.

I don't want to go on too much about this subject, but if you're currious read the wikipedia article; about 5 years ago he was charged with fraud by the SEC, for making false claims about patent ownership and illegally selling unregistered stock. A comparison of their website now and a pamphlet for the discojet from the early 70s is quite revealing; if they have made any real technological progress, it is not clear from their marketing copy!

I hope Honda is right about long-term strategy, but I'm skeptical; massive, long-term projects with private funding have not done well for their investors historically.

[Dosbomber]

If I remember correctly, that "Flying Car" (Moller Skycar) was on eBay a few days ago, going for around $3 million plus..

Yeah.. found it.. ended Oct 19th, so it was more than just a few days (how time flies... more so than this car, at least)

http://cgi.ebay.com/ebaymotors/M400X-Skycar-VTOL-Prototype-Aircraft_W0QQitemZ250036057352QQihZ015QQcategoryZ26428QQrdZ1QQcmdZViewItem

[Admin]

95 bids!?!? for a very convincing looking picture?!? 

[JonHylands]

I think it is mainly a synthesis problem.

I think we have all (or at least most) of the knowledge required, but it is scattered about, in many different companies, research labs, universities, etc.

I have spent a lot of time thinking and researching this, and have some thoughts...

In Japan, they are building some neat robots, but they all suffer (in my opinion) from the "rigid" problem - you can't build a bipedal robot without suspension for the same reason you can't build a car without suspension. You can build them, but then they can only operate in controlled environments.

I think MIT has the right idea, and in fact I think the series elastic actuators developed there and currently being marketed by Yobotics will eventually prove to be the right answer. The problem that MIT has is that most of the actual work is done by grad students, who suffer from two major problems - they are typically only involved in the project for a few years before they move on, and they typically have no real-world programming or engineering experience.

The software problem is another huge one, and I think the problem there is two-fold - again most of the people writing AI code don't have a lot of real world experience, and they are often constrained by the need to write papers for journals. They are only interested in finding "optimal" solutions, and in the real world that kind of thing often either doesn't work, or doesn't handle the boundary cases at all.

I look at video games for a perfect example of this sort of thing. Researchers in universities come up with obscure algoithms to solve hard problems, but of course they usually don't implement them. What it takes to get it done is some guy like John Carmack, who is brilliant but at the same time very practical. He reads the papers, understands the problems, and actually writes code that works, solves the problem, and integrates into the constraints posed by the whole game engine.

What you end up with is a series of game engines that are amazing, when you compare what we have now (Doom 3) with something from ten years ago (Doom 1). The information has always been there, but it takes the right kind of person to synthesize everything together, and with the development of blistering fast hardware on the side to support all this, you end up with magic.

I think we're getting there, but its going to take someone like Carmack for AI to happen. All the little bits and pieces of software and hardware are out there, but they need to be put together by a person or a group who can see the whole picture, and at the same time can actually write the code and build the components that glue everything together.

Then, we'll have the magic in robots.

I'm going to be picking up a Bioloid kit sometime in the next month or so, and I plan on starting the development of what I think are the right combination of parts and pieces to build an intelligent robot. Eventually I'll be building a biped using series elastic actuators, and integrating my control and "brain" software, and hopefully some magic will start to happen.

- Jon

[dunk]

number #2 on this list:
http://threadless.com/product/63/Damn_Scientists#zoom

dunk.
(sorry, couldn't resist.)

[Admin]

The elastic (non-rigid) actuators is a major research area in robotics now. I have met many people involved in projects trying to solve it. A major problem however is control . . . how do you acurately control something that flexes? I made a post a long time ago here asking about how to do it:
http://www.societyofrobots.com/robotforum/index.php?topic=6.0

Researchers in universities come up with obscure algoithms to solve hard problems, but of course they usually don't implement them.

This I think is a failure of the researcher to fully understand the problem . . . but what the researcher is doing is 'pure science,' understanding why things work. What an engineer does is take the science, and figures out how to make it work. Using simulations etc. pure scientists say 'a robot that does this will work well' and engineers say 'this is how you build it so it does that.' I think both working together can better solve the problems much more fully. Pure science needs grounding, and engineers needs to know why things work . . .

[JonHylands]

The neat thing about series elastic actuators is you don't need really precise control. Trying to make a leg follow an exact path is an exercise in failure, because it won't happen in the real world. Everything works in the real world (like us) with feedback loops.

When we go from a squatting position to a standing position, we start applying force to the appropriate muscles, and keep increasing that force until we start to move. At the same time our balance mechanism is providing a feedback loop in realtime that gets mixed with our perception of us accomplishing the goal (legs straightening out) to apply more or less force to each muscle group.

The same thing happens when you reach out to touch a light switch. Your arm doesn't follow a precise path, and we sure as heck don't care exactly how many millimeters away the switch is. We move our hand roughly in a path toward the switch, using visual feedback to adjust the forces being applied, until we touch the switch with our fingers.

Eventually, for tasks you repeat often, the visual feedback is not required, because you have "muscle memory". I believe that, in order for a robot to be agile the way we are, the mechanism to both learn and tune muscle memory must be the way the robot moves. I believe that muscle memory is all about the amount of force each muscle is exerting at each specific moment during the movement.

Learning motion involves sampling joint positions while being "helped" to accomplish the motion (think about how babies learn to walk), and then building force profiles to accomplish the same thing, using feedback loops from both the joint sensors and muscle force sensors.

Tuning involves subtle adjustment of force profiles to minimize the overall expenditure of force, and thus energy, combined with (for walking at least) producing a gait that results in the least amount of extraneous motion to the body (i.e., smoothness).

These theories of mine I hope to start proving first using the Bioloid kit, and later using a full blown SEA biped. It will be interesting to see how it works out...

- Jon

[Admin]

You know Jon . . . it just occured to me that I am doing this muscle memory method with my bot . . . I am fully testing the bending that occurs with a set of expected external forces with a set amount of internerally applied forces. By using statistics, I program my bot to within reason do what I want. Im basically hardcoding muscle memory, with the goal that the actuator bending will be close enough. I guess you can also call it fine tuning, or hardcoded tweaking, of elastic actuators . . . As you said Jon, I think the human brain does the same exact thing for controlling muscles . . .

[JonHylands]

Yeah, I'm pretty sure that's what happens. The only real issue is you can't just sample the forces applied to the muscles (for some configurations), because the person who is manipulating the arm/leg/whatever is overcoming gravity.

There's an interesting aside, though - if you have force-controlled actuators, you can tell them to maintain zero force, and the limb will go completely limp, and you can then grab the limb and move it as you wish, while the actuators work hard to maintain zero force by compensating for your movements.

- Jon
(edited to fix typo)