Self-Correcting VLA: Online Action Refinement via Sparse World Imagination
Classification
View four quadrants- Major category
- WAMs
- Quadrant
- Not assigned
- Architecture
- Not assigned
- Prediction paradigm
- Not assigned
- Source review status
- Not assigned
Category review. One base DiT predicts action chunks and sparse future end-effector change/progress; a separate residual controller uses those predictions and observed feedback to refine actions. Reading evidence
Contribution
SC-VLA adds progress and end-effector-change predictions to a GR00T N1.5-based flow policy, then freezes it and trains a SAC residual controller. Predicted motion supplies a directional reward whose influence decreases with predicted progress. Simulation supports both stages; physical ARX5 trials test only the predictive base policy. The evidence supports improved executed manipulation under the reported protocols, with unresolved reward and evaluation details.
Abstract
An abstract has not been added yet.
Affiliations
Tongji University; University of Technology Sydney; University of Electronic Science and Technology of China; Advanced Institute of Big Data