feat: experience-guided training with continual learning replay#98
Open
pearlq12345 wants to merge 6 commits into
Open
feat: experience-guided training with continual learning replay#98pearlq12345 wants to merge 6 commits into
pearlq12345 wants to merge 6 commits into
Conversation
added 6 commits
May 6, 2026 11:17
…checks - Validates checkpoint path exists before inference starts - Checks action_dim in config.json matches manifest follower motor count - Warns on device mismatch between checkpoint and manifest - Warns on dataset repo_id mismatch between checkpoint and current dataset - Warns if dataset codebase_version is older than v2.1 - Hooked into EmbodiedService.start_inference() - 4 new tests covering missing checkpoint, action_dim mismatch, version warning, passing case
…nferenceConfigVerifier Experience replay for continual learning was premature: - lerobot BC models train from scratch, no catastrophic forgetting - no checkpoint-resume support, so replay has no effect - confusing for research users testing different models Kept: - ExperienceStore: records training history, surfaces hints for next run - InferenceConfigVerifier: checkpoint/dataset consistency checks before inference
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What
ExperienceStore: append-only JSONL store that records training outcomes (dataset, policy, result, lesson)get_replay_datasets(): retrieves historically successful datasets for the same policy to mix into new training runsTrainSessionnow records every training submission and completion as a structured experienceTrainSessiongeneratesexperience_hintfrom past terminal-state records (success/failed/error/stopped only, not submitted)continual_learningflag (defaultfalse): when enabled, mixes historical datasets into the current training run to prevent catastrophic forgetting (ref: continual learning survey, arxiv 2302.00487)Testing
tests/test_experience_replay.pyandtests/test_train_experience.pytmp_pathfixtures