Saved in:
Bibliographic Details
Main Authors: Yu, Bin, Lian, Shijie, Lin, Xiaopeng, Shen, Zhaolong, Wei, Yuliang, Wu, Changti, Yuan, Hang, Liu, Haishan, Wang, Bailing, Huang, Cong, Chen, Kai
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.13757
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909040585474048
author Yu, Bin
Lian, Shijie
Lin, Xiaopeng
Shen, Zhaolong
Wei, Yuliang
Wu, Changti
Yuan, Hang
Liu, Haishan
Wang, Bailing
Huang, Cong
Chen, Kai
author_facet Yu, Bin
Lian, Shijie
Lin, Xiaopeng
Shen, Zhaolong
Wei, Yuliang
Wu, Changti
Yuan, Hang
Liu, Haishan
Wang, Bailing
Huang, Cong
Chen, Kai
contents Vision-Language-Action (VLA) policies are commonly trained from dense robot demonstration trajectories, often collected through teleoperation, by sampling every recorded frame as if it provided equally useful supervision. We argue that this convention creates a temporal supervision imbalance: long low-change segments dominate the training stream, while manipulation-critical transitions such as alignment, contact, grasping, and release appear only sparsely. We introduce FrameSkip, a data-layer frame selection framework that scores trajectory frames using action variation, visual-action coherence, task-progress priors, and gripper-transition preservation, then remaps training samples toward high-importance frames under a target retention ratio. Because FrameSkip operates only in the dataloader, it leaves the VLA architecture, action head, training objective, and inference procedure unchanged. Across RoboCasa-GR1, SimplerEnv, and LIBERO, FrameSkip improves the success-retention trade-off over full-frame training and simpler frame selection variants, achieving a macro-average success rate of 76.15% across the three benchmarks compared with 66.50% for full-frame training while using a compressed trajectory view that retains 20% of unique frames in the main setting.
format Preprint
id arxiv_https___arxiv_org_abs_2605_13757
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle FrameSkip: Learning from Fewer but More Informative Frames in VLA Training
Yu, Bin
Lian, Shijie
Lin, Xiaopeng
Shen, Zhaolong
Wei, Yuliang
Wu, Changti
Yuan, Hang
Liu, Haishan
Wang, Bailing
Huang, Cong
Chen, Kai
Robotics
Vision-Language-Action (VLA) policies are commonly trained from dense robot demonstration trajectories, often collected through teleoperation, by sampling every recorded frame as if it provided equally useful supervision. We argue that this convention creates a temporal supervision imbalance: long low-change segments dominate the training stream, while manipulation-critical transitions such as alignment, contact, grasping, and release appear only sparsely. We introduce FrameSkip, a data-layer frame selection framework that scores trajectory frames using action variation, visual-action coherence, task-progress priors, and gripper-transition preservation, then remaps training samples toward high-importance frames under a target retention ratio. Because FrameSkip operates only in the dataloader, it leaves the VLA architecture, action head, training objective, and inference procedure unchanged. Across RoboCasa-GR1, SimplerEnv, and LIBERO, FrameSkip improves the success-retention trade-off over full-frame training and simpler frame selection variants, achieving a macro-average success rate of 76.15% across the three benchmarks compared with 66.50% for full-frame training while using a compressed trajectory view that retains 20% of unique frames in the main setting.
title FrameSkip: Learning from Fewer but More Informative Frames in VLA Training
topic Robotics
url https://arxiv.org/abs/2605.13757