Mask loss on padded actions#894
Open
as821 wants to merge 4 commits intoPhysical-Intelligence:mainfrom
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
LeRobotDataset pads action sequences at the end of an episode out to the action horizon length using the final action in the episode. If the loss from these padded actions is applied, the policy may get stuck at the place where demonstrations typically terminate as discussed in #681.
This PR masks the flow matching loss from these padded actions in both the JAX and PyTorch implementations. DataLoaderImpl now yields a third element: action_pad_mask (or None if absent) when the policy transforms (repack, data, model transforms) pass through the “actions_is_pad” key from LeRobotDataset. LeRobotLiberoDataConfig and LiberoInputs are updated as an example.