How to play the trained agent from SKRL in Isaac Lab

[Sanjay1911]

Hello everyone,
I trained a SKRL PPO agent for my task and would like to evaluate the agent. I use Isaac Lab environment and there is a train and play script for training and evaluating the models.

However since I had to use a custom network for embedding, I adapted the skrl style isaac lab training script. However I'm not sure how I can evaluate the agent since I get configuration missing errors when I use the original play.py from isaac lab with my model.pt.

Any help/suggestion would be very helpful.

-Sanjay

[Toni-SM]

Hi @Sanjay1911

Basically you can use the code from the same file:

skrl/docs/source/examples/isaaclab/torch_lift_franka_ppo.py

Lines 122 to 133 in 90adbbc

	# # ---------------------------------------------------------
	# # comment the code above: `trainer.train()`, and...
	# # uncomment the following lines to evaluate a trained agent
	# # ---------------------------------------------------------
	# from skrl.utils.huggingface import download_model_from_huggingface
	# # download the trained agent's checkpoint from Hugging Face Hub and load it
	# path = download_model_from_huggingface("skrl/IsaacOrbit-Isaac-Lift-Franka-v0-PPO", filename="agent.pt")
	# agent.load(path)
	# # start evaluation
	# trainer.eval()

And load a trained checkpoint from persistent storage and eval the task

agent.load(PATH_TO_CHECKPOINT) 
  
# run evaluation 
trainer.eval()