swarmvolution

Will a group of neural-nets with aligned interests evolve cooperative behavior?

Demo Instructions

Run the simulator:

python demoGame.py

Run the simulation without visualization:

python demoGame.py --blind

Experiment

You can adjust hyperparameters at the top of demoGame.py

You can modify how the organisms breed, what activation functions the neural net uses, and otherwise mess with the organism code in organisms.py

DONE:

• game world
• naive predator and prey organisms
• predators eat prey
• print stats, argparse
• Implement organism vision
• have one iteration pass to detect all organism collision (instead of 4)
• move behavior decisions into the organism classes
• add organism type (PREY, PREDATOR) to organism class attributes
• fix stats summary (double counting things currently)
• implement nn brains
• implement dna
• Implement fitness functions for prey and predators
• Add control via neural-net
• Implement genetic algorithm

TODO:

• Make more parameters and hyperparameters accesible from command line
• Pickle neural-nets/universe for later use
• Refactor code to be less monolithic
• Make it easier to change placement of sensors
• Make it easier to change neural net