View Full Version : the Future?

05-11-2009, 04:59 AM
what do you think will be the future of robotics? are they still going to be the same remote controlled and repetitve entertainment toys or will there will be real humanoids that co exists with humans?

05-11-2009, 07:21 AM
Well according to an essay I had to write for class, just past 2050, robots will exceed our current population (6.7b) (based on current growth statistics). Not stating anything by that, but interesting fact.

But truthfully, I believe the toy market of robotics, in it's current state will really die, until other options become available. I mean, the biggest set back at this point when it comes to hobby bipeds, multi-pods etc, is servos. It's cheap enough to use two modified servos as motors, but when you're buying 18 servos which need to be HS-645's, it gets a tad expensive. I mean, robosapien and Wowee sorta 'robotics' is alright, but I don't know, I believe the publics expectation of robotics is going to increase a heap over the following years and the market is going to have to definately compensate for this.

We've covered most areas(computer vision, speech recognition, learning) that would accumulate to sentient robot however this has been completed individually, we have yet to combine all of these. On computer vision, i'm not sure where the video is, but it's a video of ASIMO's POV, looking at a woman who is pointing in a direction. It draws a vector from her eyes, to her fingertips to the ground, and goes there, pretty amazing actually. And on speech recognition, or well, recognition in the sense that it understands words. That British dude reteaches ASIMO what an object is, understanding his words! And for learning, you can't go past EPFL's LASA. Amazing stuff all around.

And then there's swarm robotics..

05-11-2009, 12:15 PM
I think that's a fairly generous review of AI. Sure we have speech recognizers that can get it right 70% of the time... in a lab environment. Try turning on the TV in the background, speech recognition accuracy drops like a rock. Vision is also constrained to the labs these days. Neither our cameras nor our algorithms can even get close to being as dynamic as our eyes/brains. And sure we've solved lots of navigation problems... but robots still run over people in the hallways at major research universities.

I think there are 4 major areas that will need major work before you see robots amongst humans: perception, interaction, intelligence, and run time. All must be solved in a cost effective way.

Perception - our electronic sensors just aren't there yet. LIDAR goes a long way for localization, but its really expensive. Most DARPA grand challenge mobiles had 8-15 LIDARs on them, at a whopping $20-75k a piece. No camera/algorithm out there can do many of the simple tasks we do without thinking. Consider looking at a bookshelf. There's shadows, text on books, etc, yet you have no problem telling which parts are which book... good luck making a robot do that today.

Interaction - before we can turn robots loose on the population, we need to be able to interact with them. Sure systems like SHRDLU can "understand" what you say about a blocks world, using a teletype. But we will need real time multi-user non-trained continuous speech recognition with low error rate before we can reliably tell a robot what to do (else we need to memorize specific commands). Further, we need robots that can tell us what's going on. A giant trashcan running down the hallway isn't very intuitive, hence people jumping out of the way. A giant humanoid with no expression on it's face will be just Terminatorish...

Intelligence - just having input and output doesn't get you anywhere. You need to know what to do with it. Actually, this might be one of the easier ones to accomplish, because by the time we have decent sensors, computing will be really cheap, and if we can get knowledge representation in the correct form, and are willing to settle for a task-solving robot, we are pretty much done. Then again, AI has been great at producing failures, and little else.

Run time - nobody wants a robot that needs to be plugged in every hour. If you want a life-size humanoid running around your house, well, you'll need a small nuclear reactor on board.

So where will robots be in the next 50 years? I think we'll see two trends. The cash-in-fast crowd will continue to produce things like toys and roombas. The other major trend is robots becoming us. Or rather, part of us. Since the whole see-think-act part is tough, but we have most of the mechanics, I think most of the interesting developments will be seen not in robots per se, but in robotic additions to people. Robotic arms for amputees have made leaps and bounds in the last generation. Exoskeletons that allow people to lift 20x thier weight could someday be as common as diving gear (again, run time is the main issue there). Intelligent, self-sufficient robots in labs will continue to be impressive, but won't be able to run outside the lab.. yet.

Just my pessimistic opinion....


05-12-2009, 12:59 AM
Yes. I forgot to mention that earlier. We can only do so much in an environment that the robot can work best in. The world is dynamic. Things happen without warning. You can't (as of yet) analyze a voice or face in a crowd. Neither could a robot navigate it's way through a crowd. Then robots fall, and when they're expensive like ASIMO, it's no fun.

05-12-2009, 03:46 AM
Actually, I've seen some pretty outstanding face recognition in tremendous crowds

With face isolation, regardless of the crowd size, each face is the only face.

05-13-2009, 12:16 AM
I don't know about other people but I don't think we should give robots emotions.
In the movie I,Robot, a guy says "one they robots will have dreams," but I think that kind of stuff is not neccessary. After all, robots are just to help people, not like a new species that willl share the earth with us.
Just thinking...

05-13-2009, 01:49 AM
Without some simulacra emotional response, humans won't relate to them so they can never fully integrate with our requirements.

Doctor Robotnik
05-13-2009, 02:30 AM
I don't think it has to be toy or humanoid. I think you could do quite a bit with a box with arms. It would be a lot cost effective. I think anything beyond the range of a large scale applience to car in costs will prevent a robot in every home. Even the perfect humanoid with all the perception,intelligence, ect, would take quite a while to get from lab to pick one up at the hardware store. As for emotions, They may well be given the ability to read human emotional states and thus interact with them based on it. I don't think they have to be particularly humnoid to express emotional states. The tachikomas did that quite well and look like jumping spiders. Yes I know they are fictional, but it shows you can express yourself through body language and voice.