Re: Whoa, Breakthrough!
- --- In email@example.com, Martin McKee <martinjaymckee@...> wrote:
>I know right? Of all things, the one that really bothers me is the fact that at least 3 quarters of all BEAMbots out there are either photopoppers or bicore/microcore walkers. Here's my opinion:
> I have no argument that instinct must go beyond reflexes and that it can be
> distinct from intelligence. I am not convinced that it must entail
> computation in the form of mathematics. The problem with mathematics is,
> simply, that it is precise. The world is not precise. This is the reason
> that many current approaches to AI have been so brittle. Pattern matching
> ( with memory ) is another form of computation, one which has the advantage
> ( in many cases ) of being able to generalize.
> I guess the main thing I am saying is this: I would approach ( have been
> considering ) the problem from a different viewpoint. I don't think that
> your view would be any less productive to follow however. The idea of
> implementing this as PWM in analog components will allow it to generalize
> At the same time, I think that it makes sense to continue exploring ( as
> many paths as possible in fact ). One of the main reasons that have fallen
> from the "true calling" of BEAM robotics is the lack of progress that has
> been seen in the past decade or so. The current crop of BEAM robots are
> capable enough... but they could be more so. Perhaps when I get more
> time... ( one can always dream ).
> Martin Jay McKee
NO. MORE. BICORE WALKERS. OR PHOTOPOPPERS. EVER. PERIOD.
Ok, not necessarily. It's not that they're terrible. Well, not entirely. It's just that they're so, SO numerous that they're clogging the development of BEAM in general. Sure, they're great for beginners and hobbyists, but not so good for us "serious" BEAMers. There's nothing wrong with a species becoming successful. I truly admire these simple species for their enormous success. Photopoppers and symets are extremely successful, microcore walkers were very successful, but were gradually replaced by bicore walkers. Bicore walkers are now very successful as well, not quite as common as photopoppers, but they're still everywhere. You might still see a microcore walker here and there.
But seriously, photopoppers are over two decades old now, and most two motor walkers go back 1-2 decades. You hardly see any new designs anymore. After 20 years, you'd think we'd have them complex enough that they might even copy themselves and be REALLY successful. Or actually DO something practical. But nope. All they do so far is bump into each other and fall off tables!
Well, the falling off tables thing is easy enough to deal with, but you get my point. Many bots don't even have sensors! And the rest that do often don't have enough internal complexity to match external complexity( as Mark himself would put it).
Now there ARE the select few machines that are actually innovative and adept. Those that have both the mechanical and electronic complexity to interact with their world more intuitively than the rest. 5 motor walkers are probably the best example.
And then there are those that actually perform their tasks more efficiently when simplified entirely. I cannot think of a better example for this than the amazing turbot. With only four nerves and two flagella, they can problem solve and negotiate their world both faster and more efficiently than most walkers can ever dream of. Or take uCrawler for another example. It's basically a severely devolved descendant of the 2M walker. Most would consider it degenerate. But while it's not quite as good on hard surfaces where 2M walkers excel, uCrawler actually performs outstandingly on soft or plush surfaces, like shag carpet or beds. And it also works very well on outdoor terrain, like grass, gravel, or dirt. Most of these areas many 2M walkers perform poorly in. It really depends on the particular design.
I guess what I'm really saying is that BEAMers need to either complicate or simplify their designs based on what and where the robot is intended for. Standard just doesn't cut it anymore. As if ever. The only real purposes I can think of for 2M walkers are a) turbot food, lawls, and b) experimental platforms for new and innovative technologies. Or revised takes on older technologies. But many 2M walkers simply don't belong outside the Robot Jurassic Park.
So about the actual topic, that's exactly what I was getting at, PWM calculating is just one of many paths to explore. I won't abandon the idea that calculating is an important part of neural operation, but it's only one of many important functions, and those should probably be delved into first, since calculation relies on many of these more important functions, however some rely on calculation and/or others. Therefore more complicated functions are axiomatically less important than others, yet it doesn't necessarily mean they can't be explored simultaneously. I'll certainly give you that. And generalization is a remarkable advantage. What I suppose is that both forms of computing operated in parallel would likely be ideal. Both have equal perks, and both are observed in nature, often simultaneously. So why not do the same in a robot?
- re: Suntekstore gear motorsI've ordered from this outfit; suntekstore.ca ships from Hong Kong in about 10 days to Canada. I'm happy with them and their products. Not always top quality, but good value for the price.Prices vary on the same item, creeping up a few cents each day. Something I ordered a month ago for $3.95 is now $6.78! The site will hold things in your cart for a while (30 days?) and that will hold the price too. Some prices have gone down (LED lamps for example) You have to watch for sales, too.Keep Beaming and dreamingTom