90 / 5

11th Jul 2016

5th Mar 2017

<Updated> This is a prototype for the SFPI challenge. This bot is built with a more complex brain structure; A pseudorandom relaxed signal, a sensory decipherer for wall detectors, excitable motor ganglia and a 2 min memory!

complex malloc complicated complacated basicai sfpi memory neuron brain electronics

Comments

Karakanlud 13th Jul 2016

Schmolendevice: I'm currently planning to add a second level of regulation, which would make it more complex. Adding this decoder to the memory increased the possible behavior patterns greatly. The new one could execute partial reprogramming by itself (this would need some "bad experience" input I guess). This level could also be reset this "random until optimal" way.
Killedbydeth2 13th Jul 2016

It broke itself after a little bit.
Sylvenia 12th Jul 2016

Hey Karakanlud, I am working on a light bulb that actually works like a real one, so maybe we could try to make a bot that would skrew it in? Something like, how many bots would it take to skrew in a light bulb? please reply on the sfpi page
Sylvenia 12th Jul 2016

Shmolendevice, I don't want to be mean, but that conversation is over and you're the only one that cares right now
Schmolendevice 12th Jul 2016

I also believe in fundamental goal oriented development when production of those goals through "artificial selection" is not an option. Essentially to have a neural network that in general wants to increase its sense of fullness or proximity to food whereby the neural network must train itself to optimize this process. It initially moves through randomized queries (instinct like a baby crying, even if born in a sensory deprivation tank) and strengthens connection patterns that increase fullness.
Schmolendevice 12th Jul 2016

For distance sensing, and idea I had a while ago was to implement as sort of "chemical gradient tracking" where the excitation of a neuron (firing frequency) is proportional to the distance from the target. We could at most achieve 3 level DTEC gradients for PLNT perhaps. Pretty much the bot would no longer need directional "eyes" but rather decides based on gradients.
Schmolendevice 12th Jul 2016

I'd guess that we don't necessarily need to "grow" neurons as in real life that is mostly genetically decided. On the other hand, simulating synaptogenisis and "axon guidance" on TPT is what's difficult.
Sylvenia 12th Jul 2016

Something like, how many bots would it take to skrew in a light bulb?
Sylvenia 12th Jul 2016

k thanks, I am working on a light bulb that actually works like a real one, so maybe we could try to make a bot that would skrew it in?
Karakanlud 12th Jul 2016

Sylvenia: Of course.

The Powder Toy

My Tiny Brain 2 by Karakanlud

Comments