Issues with multi-agent task's quick start commands

[jinwoolee1230]

Dear maintainers,

I really appreciate your simulator and I hope I could use this as a research platform.
However, I could find some issues related to multi-agent tasks.

1. Platform hover quick start command: python train.py task=PlatformHover algo=mappo headless=true eval_interval=300 total_frames=150_000_000 does not work but it works when the algo=ppo. I am not so sure is there going to be any issue or not but I think this is not a major issue though.
   I get Error executing job with overrides: ['task=Formation', 'algo=mappo', 'headless=false', 'eval_interval=300', 'total_frames=150_000_000'] RuntimeError: TensorDictModule failed with operation error
2. Formation task does not work neither in algo=ppo nor algo=mappo.
   I get Error executing job with overrides: ['task=Formation', 'algo=ppo', 'headless=false', 'eval_interval=300', 'total_frames=150_000_000'] RuntimeError: TensorDictModule failed with operation error

Can you please give me some instructions about these issues?

Thank you so much for your kind help and instructions.

[btx0424]

Sorry for the trouble! I refactored MAPPO implementation to be cleaner but left out the handling of complex nested observations (using a transformer) in some mult-agent environments. Will fix that soon!

[mrtbrnz]

Hello, I also want to thank you for your work and making it open for us ! Appreciated a lot !
Likewise I managed to get the hover and track environments work for single agent, but couldnt get any of the multi-agent envs work.

Could you please give an example multi-agent task and algo couple that works with the current repository ?
Thanks !