CarGame is slowly expanding. I plan to release update at the end of this month.
For now, Car8 is on the wheels and a river has been added to the environment.
I have tried before to make 3D spheres steer towards a blue orb (arbitrary point in space) with a genetic algorithm (of neural networks), but it never quite worked out:
I eventually decided my approach was wrong, and that I should give them some sort of ‘eyes’ and a less forgiving natural selection, but I have not yet created something new.
How are your creatures evolving? Do you mutate some every generation or are they learning on the go?
Also, how do you make them steer? Do they give a direction of output or a rotation (do they strafe or do they rotate)?
I’m interested in your aproach, because it looks really good.
How are your creatures evolving? Do you mutate some every generation or are they learning on the go?
Also, how do you make them steer? Do they give a direction of output or a rotation (do they strafe or do they rotate)?
Well let me try to briefly explain what I got so far:
Every creature as you might see has green parts, and each green part has inputs/neurons/and outputs. Each part could be connected to other creature parts as well.
The output could be a central force applied by a vector defined on each part…or also a torque with the direction of a vector too.
The population is for instance 10 creatures, and on each generation I took the best ones (lets say I took 5) and fill the remaining population with a cloned version of those best ones, but with some slight mutation on the weights of the neurons - and evaluate again and again…
I did many tests but I still don´t see a good result…I suspect my evaluation function is the problem but Im having so much fun.
Anyways, I could share the code if you like, just ping me.
Using my new Lemur MigLayout class, my new anchored pullout container class and my new rotatable text component class I can now implement animated pullout panels in minutes.