RL-based Multi-Agent Simulation for Crowd and Predator-Prey Behaviours

[kayuksel]

Hello, I am training the mentioned behaviors with RL on a simulation environment I wrote in PyTorch.
The videos of the learned behaviors are as follows, especially the prey-predator one is quite interesting.

Predators that are learning to hunt their preys: https://www.youtube.com/watch?v=Ds9O9wPyF8g
Crowds moving to randomly assigned goals: https://www.youtube.com/watch?v=VC5WT70MBQU
Crowds moving to the opposite side of a circle: https://www.youtube.com/watch?v=B48TkaBFe-c

It would be great if we could port these to Brax. I never used Jax but previously used Brax PyTorch API.
I am writing to see if I can find anybody to collaborate/help. I also started creating a Gym environment.

Source codes for videos shared above can be found here: https://github.com/kayuksel/multi-rl-crowd-sim