What is your evaluation strategy?

Hi guys, I’ve been using Isaac Gym to work on robots. To be more specific, I’m focusing on legged gym.

Currently I’m working headlessly on a server, and I record the camera video for debugging. I’m wondering what’s the best/proper way to evaluate the policy in Isaac Gym? For example, I have 10 parallel envrionment for training (let’s call it env-10), but I am not allowed to create a only 3-env parallel Isaac Gym environment (env-3) with different seeding for evaluation. Do you have any good strategy for evaluation? I want the eval env to have another seeding and can “step” separately.

Thanks!

1 Like

Hey! I’m wondering the same thing; did you ever figure out a good way to do this?