ICOR (Robot collab)

[savask]

We don't need to make a special sensor for shock. We can use mine pressure sensor, for example when it will give very big value, spinal chord will think that it is a shock. But it isn't easy task to make a spinal chord, so in my first robot I used a special key for teaching.

[Vytas]

I think we will need a bit more experience in unsupervised neural networks. I think that's the key to the spinal chord teaching. I also have been thinking of a GA (genetic algorithm for teaching), but I wouldn't know how to setup the simulation and I have no experience in GA's whatsoever.

[savask]

I'll think about unsupervised neural networks later, but now I should make a computer vision for robot.

[Vytas]

That would be great!

I have a feeling that it would involve lasers am I right?

[Bronie12345]

i just got an idea for something to teach robots!!!
too bad you need the human to give some examples.
so, when you hold return (enter), it goes to learning mode.
then press a key (assigned to a movement), and it will ad that movement to the state of all buttons.

like, a robot has 4 sensors
when it is at all negative, the variable it'll read = scene.my.0000
when let's say, the left wall detector is positive, it will read: scene.my.1000
etc...
how the code will be build up to read: {"scene.my."+scene.my.swl+scene.my.swr+scene.my.swhl+scene.my.swhr}
that's how i though something to teach a robot

[Vytas]

Well that would work for a wall detector, but not for a learning robot.

[thebeech]

why not make it have like a virtual GPS system. it takes the e.pos of an object and moves along a path for example a maze, then if it hits an obstacle then it traces back and erases the track from its system, so it basically just builds a map of an area with a path that it saves.

[Bronie12345]

with my idea, it is kindof a long work, but it is worth the wait.
you can also add robotic arms, and more, and them press a key when you're in learning mode, and when all sensors are that way, it always does the job you gave it.

for newcomers it will be simple programming with some kind of IF command

EDIT:
i'm now figuring out the writing part of the memory.
i already know how to do the reading part.
oh, and all the sensor data is stored into only one variable...
only simple/basic scripting is used. (so no math functions etc)

[Nait]

Sorry if i'm not in time, i have problems with internet.I,ve got an idea : robot has eight sensors, two on every side, it is going until it meets something, then it detects from what side it is situated, looks in memory, memory of robot can be an 3D or 4D array like this [[a,[b,c,d...]],[]], a is number of placement it is counted as sum of numbers of sensors, b - what action are made exm.
1 - go backwards,
2 - go forwards,
3 - turn on flying device,
something like this.
and it doesnt find combination needed, it makes a enumeration of possibilities and write every step (action) in random memory. If it is good variant and robot can move forwards it writes in HD new square, in it number of plecement and actins from RAM. If this possibilite isn't good and robot checks sensor again it makes all steps in back order, and reset RAM, and tryes again.
what do you think of this ?
If my english isn't good I'm not guilty.

[Bronie12345]

ill first finish my own example, ok?
you think you can make something with that, because you're also going to need to work in binary etc.

we don't have RAM yet, but i think that won't be too hard to make.
i also just got an idea for the RAM setup, but that will take a while to finish.
the RAM will be portable so you can use it in diffrent devices.
it will also be easy to control...

[savask]

2 Vytas
Yes, you are right. First camera prototype will return a distance up to the object.

2 Bronie12345, Nait, thebeech
As I see, there are 2 groups of people in this collab. The 1st group thinks that we should start from neural algorithms (me and maybe Vytas), and the second group thinks that we should make a robot with different position arrays, moves storing system and etc. So I decided to make a kind of competition. You should prove that your idea is better than our, and we will prove our idea too.

I think we will find a compromise

[Nait]

Sorry, Bronie12345, I was really wrong, lets finish your idea at first. But I'll make my model ! I've got a new idea : robot exists in a space (I mean not outer space) until some kind of sensor or detector activates and sends nerve impulse to spinal chord by-turn it stops all movements of robot and turn on teathing mode.
Let's remember us in childhood if we see something new "Oh, what is this", and you'll explore it.
So next step will work with different sensors, robot will collect information of it. Idea with e.pos was good, at first robot will write all things he meet than in normal mode it is condition "There was something in [cordinates], am I near it"."for i := 1 to n do if e.pos = memory[i]+-10 than turn on danger mode".Oh I can't forget Pascal codes, realy iI know how to write right. I thik that to manage without
some kind of memory, arrays or not.If you want you can make a lot of neurals or logic gates, becouse they are working alike.In danger mode it it makes reverse and finds ways to go across obstacle.
Oh I became tired while typing all this . Savask тебе персональный привет из Владика

[Bronie12345]

can't wait until i have internet again (im now at school)
so, for a while, it'll take me some time to get it finished, and also will take some time to upload...
i really need my laptop to be 'repaired', so i can do everything on my laptop incase of a crappy school computer with *** limitations.

on-topic again:
first, ill make the robot, then the RAM, then a robot that uses the ram.
and the ram has an infinite amount of storage, because of the way of saving the stuff.

[Vytas]

Well I'm definitely sticking with neural networks, simply because the infinity state model as referred by alot of people at this topic sucks. Sure it is easy to implement and to code, but it hasn't got the same possibilities as a neural network. Maybe you guys should do some research? And with an infinity state model each pattern should be hard coded, but with a neural network it learns to do it's own patterns.

[Bronie12345]

with hard coding, it can also learn it's own patterns.
you can simulate a neural pattern with hard-coding.
mine will be a bit diffrent, but will still be able to learn.
though, i need alot of time to make these kind of codes.
and it WILL be phunlet-safe

[Vytas]

Well if your simulating neural networks you aren't using the infinity state model.

[Vytas]

So your with me and savask?

[savask]

Neural programming is good for situations when we don't know what algorithm robot should use.
And it have one big "+" :the theorem of convergence of a perceptron (I don't know how it will be in English), this mean that it will solve a problem anyway. But we will need to make more layer perceptrons to make it to solve harder problems.

2 Nait
Здорово что здесь есть ещё Русские

[Vytas]

Yes that's correct.

Savask I'm going to use your neural network as a base for an unsupervised neural network I was talking about. I'm guessing this ain't going to be a walk in the park, so it could take a day or two too get it working. And then maybe set a few required outputs and let the robot learn. Then "maybe" we could persuade the others to get working on neural networks as well.

[savask]

Oh, don't use it now,please, I have a better one. I'll present it when I make my camera work.

[Vytas]

Ok, wasn't busy on it yet first I'm building a complex one in javascript not in thyme because I'm not real good at it.

[Nait]

Savask, а perceptron, ты имел в виду устройство которое распознает обьекты или что?
И еще у меня есть книга думаю пригодится. Называется "Юный кибернетик", автор Ю.М. Отряшенков 1978 г. У меня есть но как гиперссылку сделать не знаю.

[savask]

Вот хорошая статья из Википедии.

[Nait]

Спасибо, а книга тут. А DJVU-reader тут.

[Bronie12345]

please in english again