Hi, thanks for your nice work!!
I'm wondering if you have any code implementation to calculate epic distance between ground truth reward function and learned reward function??
Concretely, I have no idea how to get true reward from uniformly sampled state, action pair.
How can it possible to calculate reward from certain state-action pair in metaworld environment??
It gonna be possible if the state was sampled by env.reset() and the reward was calculated by env.step(). but as far as I know, it's not available calculate reward from user-defined state.
Best regards,
Jongkook, Heo
Hi, thanks for your nice work!!
I'm wondering if you have any code implementation to calculate epic distance between ground truth reward function and learned reward function??
Concretely, I have no idea how to get true reward from uniformly sampled state, action pair.
How can it possible to calculate reward from certain state-action pair in metaworld environment??
It gonna be possible if the state was sampled by env.reset() and the reward was calculated by env.step(). but as far as I know, it's not available calculate reward from user-defined state.
Best regards,
Jongkook, Heo