add a new environment and agent class for "isolation"
The configuration of an "isolation" simulation is quite annoying atm, because the configuration for the model is not really made for that kind of simulation. This should be changed to a new environment and corresponding agent class.
environment class that contains:
- (small) grid size
- bigger viewing range
- potentially coloured input for agents to better distinguish between the agent itself and its surroundings
- fewer actions:
- 5x move
- 5x eat
- 1x procreate ? (random placement for offspring if any cell in the 9 neighbourhood is free?)
- or even completely differnt actions? (look for inspiriation in nature?)
agent class that contains:
- its very own neural network (completely configureable via config file)
- potentially 2 inputs; 1 convolutional for neighbourhood, 1 for food reserve?
- adapt the outputs corresponding to the agents' action space
- "classical" properties like
Overall rething the