Privileged Foresight Distillation: Zero-Cost Future Correction for World Action Models
Classification
View four quadrants- Major category
- WAMs
- Quadrant
- Not assigned
- Architecture
- Not assigned
- Prediction paradigm
- Not assigned
- Subcategories
- Efficient inference & real-time control
- Source review status
- Not assigned
Category review. Future-enabled and current-only masks share a WAM backbone; privileged residual distillation transfers future-conditioned action corrections into action-only inference. It modifies a complete world-action policy, not a generic distillation building block. Reading evidence
Contribution
Privileged Foresight Distillation (PFD) compares two attention masks on the same world-action backbone, then learns the future-enabled change in action velocity through an output adapter. Deployment uses only the current frame and corrected action denoising. The reported LIBERO mean improves by 1.15 percentage points over reproduced Fast-WAM, with a measured 5.15% cached-context latency overhead; the title's zero-cost wording does not mean zero additional computation.
Abstract
An abstract has not been added yet.
Affiliations
The University of Southampton; The University of Queensland