Getting an error when using TD3 in parallel setting OmniverseISAAC Gym

[dhruvkm2402]

Hello,
I'm getting this error when trying to use TD3
File "tasks/hunter_train2.py", line 116, in
trainer.train()
File "/home/krovilab/.local/share/ov/pkg/isaac_sim-2022.2.0/kit/python/lib/python3.7/site-packages/skrl/trainers/torch/parallel.py", line 141, in train
self.single_agent_train()
File "/home/krovilab/.local/share/ov/pkg/isaac_sim-2022.2.0/kit/python/lib/python3.7/site-packages/skrl/trainers/torch/base.py", line 171, in single_agent_train
actions, _, _ = self.agents.act(states, timestep=timestep, timesteps=self.timesteps)
File "/home/krovilab/.local/share/ov/pkg/isaac_sim-2022.2.0/kit/python/lib/python3.7/site-packages/skrl/agents/torch/td3/td3.py", line 213, in act
actions = self.policy.act(states, taken_actions=None, role="policy")
File "/home/krovilab/.local/share/ov/pkg/isaac_sim-2022.2.0/kit/python/lib/python3.7/site-packages/skrl/models/torch/deterministic.py", line 91, in act
taken_actions.to(self.device) if taken_actions is not None else taken_actions, role)
TypeError: compute() takes 3 positional arguments but 4 were given
Exception ignored in: <function _make_registry.._Registry.del at 0x7f9ddf57fb90>

I was looking at the deterministic.py files and trying to debug but the same error. What modifications should I do to use TD3?

[Toni-SM]

Hi @dhruvkm2402

It seems to be one, the main script or the installed version of skrl is out of date.
This error is related to the way computation is defined within the models for previous skrl versions (<0.8.0).

Two solutions:

• Check the skrl version (pip show skrl) and install the latest version if it is out of date
• Update the .compute(...) model method by looking at model definition or examples in the latest documentation page

Please, let me know if everything goes well after the changes

[dhruvkm2402]

Hi @Toni-SM
I updated skrl to the latest version and it works now. Thank you for your help!