Simple Predator-Prey Simulation

I wanted to get some experience with this kind of things (Reinforcement Learning), and I once sam some sililar exeriment somewhere, so I tried it too.

Setup

In the simulation there are two prey agents (green & blue circles), trying to get more XP. In the arena there are also randomly distributed red an greend dots. The prey will get 1 xp for "eating" a green dot, and lose 0.5 for a red one. Both have their own NN that is trained during the simulation.

There is additionally an slow predator (red dot), always trying to get to the closest prey. On collision the prey loses some XP.

Goal

First of all, I want to learn more about RL, of course. But I am also interested in the behaviour the prey agents will develop. They could potentially develop team-strategies, to escape the predator agent.