ICOR (Robot collab)

[Bronie12345]

when you start the scene, his memory is empty, but he learns when he does things
like,in every situation it tries stuff, but it saves the move when it is moving again

[savask]

I'll explain you my idea.
There will be spinal cord and brain.
Spinal cord will store congenital reflexes, and robot can't change them.
And brain will store changeable reflexes (Classical conditioning), and robot will get them from real life like Pavlov’s dogs

[Vytas]

I have an idea for a more physical neural network, which wouldn't involve lots of scripting, working on that right now. Maybe tonight I have something that I could show you guys.

[savask]

It will be interesting to see

[Vytas]

I have a question regarding the robot. How are you going to get the input signals from the environment? lets say you want the robot to pick up all green boxes how would the the robot "see" the green box.

[savask]

There will be a kind of sensory receptor. It will give the robot a velocity of object touching it, and I have some prototypes of this kind of sensor

[Bronie12345]

use touch sensors, pressure sensors, and color&touch sensors

Code: Select all

onCollide sensors:

touch sensor:
onCollide:   (e)=>{scene.my.sensor1=true}

color and touch sensor:
oncollide:    (e)=>{scene.my.sensor2=true ; scene.my.senseCC2=e.other.color}


LaserHit sensors:

hit sensor:
onLaserHit:    (e)=>{scene.my.lsensor1=true}

color and hit sensor:
onLaserHit:     (e)=>{scene.my.lsensor2=true ; scene.my.lsenseCC2=e.geom.color}


[savask]

No, your pressure sensor is bad. It can't return velocity (pressure) of the object.

[Bronie12345]

ok, i've edited the code

[oops_ur_dead]

Well I guess you COULD have basic learning, such as the bot knowing where obstacles are and changing variables based on past "experiences", but we cannot make the bot do any complex actions (such as movement patterns) that we did not explicitly program it to do without hard work. It could be possible by creating a thyme scripting machine with a complex AI that can modify the script of the bot's "brain", but that would be astonishingly hard.

[thebeech]

i would like to join helping with anything scripted or mechanical

[oops_ur_dead]

I have an idea on how this bot could "learn". There are four layers of processing
Input (sensors)
Brain
Cortex
Output (engine)
Data comes in through the input. The input will set variables that the brain then reads. The brain controls the processing of the data based on constants and variables. The brain will change two sets of variables. The output variables and the learning variables. The cortex will read the learning variables and change certain variables that the brain reads to make decisions. For instance, if the brain receives lots of impact inputs, it will record these. The cortex will make the decision that the area is dangerous, and record the co-ordinates of the area in variables. The next time the brain enters this area, it will either be cautious or try to leave.

[Bronie12345]

my idea of learning was:
1. it gets in a situation where it is stuck (can't move or something)
2. it's going to try how to get out of that situation
3. when it gets out, it saves the movements it made.
4. if it gets into that situation an other time, it does that actions over so it gets out.
5. when it gets out faster with diffrent movements, it saves that and does that next time

ill try to make a robot that can remember some stuff
(like, it can save if a situation is dangerous or not)

[oops_ur_dead]

Thats possible. The problem is that it will be hard to distinguish between what movements helped the robot get out of its positions, and which movements did not do much. We want the robot to be as efficient as possible, so it would be undesirable to save the movements that didn't help the situation. Otherwise, it is a good idea.

[savask]

You guys should read this and this.

My idea of learning:
1. When something occurs often, robot saves it to the special list like reflex.
2. When something occurs, robot search reflex in the list.
3. If reflex is unused long time, it will be deleted.

[Bronie12345]

like a short-period memory?

[savask]

No, it will store information for long time.

[Bronie12345]

ok, but when its not used, it 'forgets' the data
something like that?

my way of saving is diffrent...
it just uses the sensors to make a variable to save...
like, you have 6 sensors...
when all sensors are 'false', the storage code is: [0, 0, 0, 0, 0, 0]
when sensor number 4 becomes 'true', the storage code will be: [0, 0, 0, 1, 0, 0]
that's how it creates a variable for everything that could happend...
that's how i was thinking to do it...
and ill also put an accelerometer into it...

[savask]

It forgets the data only when it isn't used for a LONG time. It is like your brain work.

[Sniperkasa]

Hmm.. if I may.. I'd like to join this collab ;D
Got one good idea wich I will post once i get it done

[savask]

Would you like to have a status, for example mechanic or neural programmer?

[oops_ur_dead]

You guys should read this and this.

My idea of learning:
1. When something occurs often, robot saves it to the special list like reflex.
2. When something occurs, robot search reflex in the list.
3. If reflex is unused long time, it will be deleted.

That is good, but as I said above it would be hard to distinguish between what moves helped and what moves did not.
For the robot, I think that we should have more sensors. Im working on an accelometer. The current contact sensors we have are kind of slow, because the robot has to physically touch the object to sense it. I suggest we replace it with lasers.

[Bronie12345]

an accelero-meter isn't that hard to build
but then it also measures the speed, so thats an exra

[savask]

I think that we should make a kind of artificial neural network. And then, we will optimize it replacing neurons with thyme scripts.

And, yes, about sensors. We should have a pressure sensor, because robot can understand
what is dangerous and what not.

[Bronie12345]

that's a nice way to go, but a little above my scripting lvl...
anyway, im crrently working on something else (yes, also alot of scripting)