[HS]

I'm trying to think of a circuit for this:



The main idea behind the system is to narrow in on something. Start with a few cubes in a very simple robot, subjected to real life sensory inputs. Make random (if that's a logical concept lol) LED's light up in response to inputs until a good thing happens. Have it incorporate this into it's memory, build on that, keep going. Eventually optimal outputs for various inputs will be discovered and kept. Like a genetic algorithm making those stick creatures which move across to the right. The are lots of demonstrations of this process on YouTube. The real neurons will have to be smaller, but if I can make this large scale model work, then I'll know if it's worth pursuing on a more impossible small scale

Here is the circuit I came up with:
BLUE = SOLAR CELL
YELLOW = DIFFERENTIAL OP AMP
GREEN = MICRO-CONTROLLER
RED TRIANGLES = LED'S

Does it look like it would work? And, can you think of a better way to do the same thing?

[ranch vermin]

How are u going to connect your cells together -  werent you going to use a transmittive system?  that was the big win for doing it this way right?
Im not sure what the make of the circuit yet,  an explanation on the side would be nice - like - what is it what you want it to do is the main thing.


With evolution,  you have so many trial connections,  you give each connection type an experimental time period, and then you pick the best one,  then you start again.

So these trial connections - are going to be different synapse weights on each "frequency"  depends on how you do it.

Then once its found the best, due to a rule,   then it just keeps going until it hits its optimum.

Doing it this way, your avoiding putting the mesh of wires in each level.   its a good idea i think... keep me posted!

[HS]

Transmission happens when solar cells on one cube register a blinking light coming from a different cube.
There will also be a charging circuit connected to the battery, 6 solar cells in series with a check valve diode. When the charging lights turn on, the sum of micro-controller input voltages will be high, this will cause the signaling circuit (what is shown in the diagram) to go into a wait loop (for example, wait 10 seconds, check voltage sum, loop.) Once the charging lights turn off, the signaling circuit will "wake up" and continue improving the conectome till the batteries run low. Thanks for clarifying the evolution method some more.

[ranch vermin]

Ive got an idea for ya.     [EDIT] unless u were already doing this?? [/EDIT]

So your reading a light blink,  one thing you could do is time the blinks, and depending on the time of the blink is what cell it came from?

If you did that,  it probably might not be the best solution, you would have to slow the system down to the amount of cells.   so if it were a billion cells,  youd have to run at a gigahert to have a full connection in a second...  but then it would still be wireless communication.

Also,  youd be limited to the frequency an led can turn on and off,  but i think from what ive read they are quicker than the eye by a huge margin.


[ed]a better solution would accept the signals in parallel - then it would be over in a flash.

[HS]

Oh yeah, that's pretty good! If it can identify cells, it can learn which cells to accept signals from. Like a noise filter, only useful signals would have to be processed. Of course once in a while the cell would try new connections, if  there is a need to encode new information, like if it revives a foreign signal. Then it would try to make new connections.  But otherwise keep it fast.

[ranch vermin]

you could do it with a 555 timer - hooked up to an incrementing clock.   u could get ic's for both.   Then each cells "flash position" is one step of the clock.

Then I guess your system is reading a slight difference in lighting amplitude, each clock step.   Then u have the weights stored in the cell,   and if it peaks the threshold then you are in fire mode.  and then other cells are checking for its clock position - to multiply by their weights, if they fire, etcetera..

[ed]  just thought of a prob tho,  all the timers would need to be in synch with each other.  so i wonder if you have a green led, thats used to synch their clocks every cycle of the clock?

[HS]

If I got some colour detecting diodes, the green led thing could work. I recently spotted a deal and bought 4000 [Edit] 2'000 [Edit/] led's  so I've got enough for the foreseeable future. I would like to have each cell running at full speed, so I'd need a cell detection system that doesn't take up any time. Maybe a light source triangulation. Then each light source could get a name based on it's position. Like a star map for telescopes.

[ranch vermin]

That sounds interesting,  theres no limit until you attempt to do something impossible, but thats a good part of science and engineering!!   

https://electronics.stackexchange.com/questions/70926/led-shut-off-time  <-im not sure about how factual it is,  maybe look up "led filmed with high speed camera"?

[ed]  a perceptron can actually run (be setup) like a computer as well,  if u finish this uve made ur own computer!! [/ed]

[HS]

Yeah, I guess this could be set up in an orderly fashion and do calculations in a traditional way. Chaos computers FTW though