Robotics & manufacturingResearch findings
SEAM: A Smooth Execution Method for Action-Chunked Vision-Language-Action Policies
Original published 2026.07.06 11:32 KSTAdded to Blackdeck 2026.11.22 04:45 KST
What has changed?
By leveraging the unexecuted tail of the previous action chunk, SEAM enables smooth transitions between action chunks without policy retraining.
Key points
- Vision-language-action policies exhibit multimodal bifurcation between action chunks, causing abrupt discontinuities.
- SEAM uses the unexecuted tail of the previous chunk as a consistency reference during inference.
- Boundary jerk is reduced by 28%, transition discontinuity by 27%, while maintaining task success and denoising cost.
Connect it to your work
Achieves smoother motion transitions in action-chunked policies without backpropagation or retraining, reducing boundary discontinuities by 28%.
What remains unproven
This research note is based on an abstract; check the original for test conditions and application scope.
Originals used for this note
Links open the original publisher in a new tab.