proroklab / VectorizedMultiAgentSimulator

VMAS is a vectorized differentiable simulator designed for efficient Multi-Agent Reinforcement Learning benchmarking. It is comprised of a vectorized 2D physics engine written in PyTorch and a set of challenging multi-robot scenarios. Additional scenarios can be implemented through a simple and modular interface.
https://vmas.readthedocs.io
GNU General Public License v3.0
350 stars 71 forks source link

movable target flocking #106

Open armansouri9 opened 5 months ago

armansouri9 commented 5 months ago

Hi. I want to train flocking scenario with fixed target in some position.

1) How to false movable target in my training? I have this code and I want to use movable target in my .yaml file like your demo of flocking.

https://github.com/pytorch/rl/tree/main/sota-implementations/multiagent and another request:

2) can you send your flocking env hyper parameters for trained model in your .yaml file that you made demo on youtube?

Thanks.

matteobettini commented 5 months ago

Hello,

What video are you referring to?

We currently do not have an option for training flocking with a fixed target.

The hyparparameters I would start from for flocking are those in https://github.com/pytorch/rl/blob/main/sota-implementations/multiagent/mappo_ippo.yaml and https://github.com/facebookresearch/BenchMARL/blob/main/fine_tuned/vmas/conf/config.yaml