Hi, > Sorry for the delay, I think this is a very good offer. I was busy > trying to figure out a way to accelerate neural processing, but it > seems not so easy.
Well if you figure it out, let me know. Although my neural processing seems fast compared to that of the average American, it often seems frustratingly slow to me, and I'd really like to have a way to speed it up ;-) > Anyways, I took a look at SmartRobots.com, their models don't have a > visual (camera) module. Sensors are limited to obstacle avoidance. > It may be possible to add a visual module to it, In fact they are trying to convince people to create a nice visual module for it; see the following: http://www.smartrobots.com/call_to_arms.php > but my original idea > is to develop a minimalist visual module without robotics, for the > simple reason of cost cutting. If we were to develop a full model, > the product will be more expensive and time-to-market will be longer. > I guess the best approach is to develop a visual module that can be > added to varieties of robotic platforms. While it's certainly possible to do interesting research on "vision processing in isolation", I think it's more promising to work on vision in a context of -- multisensory integration -- integration of perception with action To take a simple example, understanding how to recognize a chair is easier if you can move around and look at the chair from a different direction, roll the chair over and see what it looks like from the bottom, etc. Having a system learn object-recognition in isolation is surely possible, but IMO makes the learning problem harder rather than easier. > The AGI flatform will be one based on experiential learning, as Pei > Wang suggested earlier. Well, I like to talk about Experiential Interactive Learning -- and it sounds like what you're proposing is missing the "interactive" part ;-) > > I'm worried that some robotics companies may be developing their > platforms as well, and there'll be issues of standards, compatibility > etc. > No question there -- but the problem of making object recognition work robustly is obviously a lot harder than the problem of translating between different formats or different so-called "standards" ... Make object recognition really work, and the world will make your own formats "standards" ;-) -- ben g ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
