Multi-Agent RL Environment for CrowdSim, Predator-Prey, and Army

[kayuksel]

Hello again,

I created a crowd and predator-prey environment that may be useful for ejovax. I also work now in a melee combat environment.

Here is the PyTorch implementation of the crowd and predator environments: https://github.com/kayuksel/multi-rl-crowd-sim

Here is a video where multi-agent predators are learning to surround preys to maximize hunts: https://youtu.be/Ds9O9wPyF8g

(I will also create a competitive multi-agent environment for closed-market auction where they will self-play by placing orders).

Have a nice week.

Sincerely, Kamer

[kayuksel]

I have also just seen @RobertTLange's notebook examples on optimizing multi-agent boids for flocking. Great stuff! I would be glad if we can collaborate on the melee-combat battlefield self-play, which turned out to be quite complex. (In predator-prey or melee-combat, there are two groups competing in a game-theoretical relationship between them.)

[lerrytang]

Hi, thanks for sharing the demo, it's very cool! We are open to collaborations. Also, if you are interested, we would love to have your PR to include this crowd and predator demo in evojax.