rsl_rl reproducability

[SourabhPrasad]

Hi,
I'm working on training a quadruped robot using a modified version of rsl_rl that includes an auxiliary MLP network.
I've encountered a problem with reproducibility: even when seed is explicitly set while executing the training script.
Do I need to manually set the seed anywhere else in addition to DirectRL's torch_utils.set_seed()

[RandomOakForest]

Thanks for posting this. Using --seed when launching train.py should allow you to reproduce a model.

[SourabhPrasad]

Hey @RandomOakForest,
Thanks for the reply. The --seed=42 was set while executing the training script.