Binary Env results? about corl

[nakamotoo]

Adroit-binary datasets are not included in the original d4rl dataset. They are collected in AWAC's paper: https://github.com/rail-berkeley/rlkit/tree/master/examples/awac

Here are several recent works that utilize the binary dataset that might be helpful:
Cal-QL: https://github.com/nakamotoo/Cal-QL#adroit-binary
RLPD: https://github.com/ikostrikov/rlpd/tree/main#adroit-binary

[DT6A]

As far as I remember, binary datasets are relabeled and cutted d4rl "cloned" datasets. "cloned" datasets are "human" datasets + trajectories sampled from some BC policy. This is why we benchmark on cloned datasets

[DT6A]

Hello. No we don't provide results for the binary Adroit datasets. You can check binary datasets usage here: https://github.com/nakamotoo/Cal-QL/blob/f3f9bcc853ed5df96b2f6cd4b9e4d84d5ba0b046/JaxCQL/replay_buffer.py#L125

[StoneT2000]

Maybe @nakamotoo can help clarify?

Was wondering where does this binary demo data come from? Is it the original D4RL dataset but relabelled to have sparse reward?

And where is this BC data from?