Isaac Lab Multi-Agent skrl documentation not working

[Saumya-Karan]

I tried running MAPPO to train the multi-agent environment Isaac-Shadow-Hand-Over-Direct-v0 with skrl using the following command:
isaaclab.bat -p source\standalone\workflows\skrl\train.py --task Isaac-Shadow-Hand-Over-Direct-v0 --headless --algorithm MAPPO

I am getting the following error:

Error executing job with overrides: 
Traceback (most recent call last):
File “C:\Users\saumy\IsaacLabrepo\IsaacLab\source\extensions\omni.isaac.lab_tasks\omni\isaac\lab_tasks\utils\hydra.py”, line 99, in hydra_main
func(env_cfg, agent_cfg, *args, **kwargs)
File “C:\Users\saumy\IsaacLabrepo\IsaacLab\source\standalone\workflows\skrl\train.py”, line 178, in main
runner.run()
File “C:\Users\saumy\isaaclab\lib\site-packages\skrl\utils\runner\torch\runner.py”, line 376, in run
self._trainer.train()
File “C:\Users\saumy\isaaclab\lib\site-packages\skrl\trainers\torch\sequential.py”, line 84, in train
self.multi_agent_train()
File “C:\Users\saumy\isaaclab\lib\site-packages\skrl\trainers\torch\base.py”, line 293, in multi_agent_train
shared_states = self.env.state()
File “C:\Users\saumy\isaaclab\lib\site-packages\skrl\envs\wrappers\torch\isaaclab_envs.py”, line 136, in state
return self._env.state()
AttributeError: ‘OrderEnforcing’ object has no attribute ‘state’

Set the environment variable HYDRA_FULL_ERROR=1 for a complete stack trace.
2024-12-13 07:46:35 [14,426ms] [Warning] [carb] Recursive unloadAllPlugins() detected!

Please explain how to resolve.

[Toni-SM]

Hi @Saumya-Karan

This is because gymnasium-v1.0.0 deprecated direct access to the DirectMARLEnv.state method.

I am going to update next skrl release to take it into account.
Meanwhile, please use gymnasium-v0.29.1: isaaclab.bat -p -m pip install "gymnasium==0.29.1"

[Rishi-V]

I encountered a similar issue using --algorithm=IPPO.
When I looked at the Trainer.multi_agent_train(self) function in skrl/trainers/torch/base.py I saw that the only places that env.state() are used are for the infos dictionary. So my hacky current solution was to comment out the few lines referring to env.state() and shared_states and I am able to get IPPO running.

I plan to create several custom MARL environments with Isaac Lab this new year (Happy New Years!) so looking forward to an official fix. 🙏

PS: Thanks for writing this codebase in the first place! 🙂