Diffusion Policy — RoboCasa Activations (latest checkpoint) Per step, per episode activation traces collected from a DiffusionTransformerHybridImagePolicy (the diffusion policy library) rolled out on RoboCasa benchmark tasks. Captured with collect activations robocasa.py at the latest training checkpoint. These traces are the input expected by the conceptor / SAE steering pipelines under diffusion policy/experiments/robocasa steering/ and diffusion policy/experiments/sae/ — see… See the full description on the dataset page: https://huggingface.co/datasets/brandonyang/diffusion policy robocasa activations latest chkpt.
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy