How Many Demonstrations Does a VLA Need? On the Sample Efficiency of VLA Finetuning
Overview
This research investigates the data requirements for adapting pretrained vision language action (VLA) models to new tasks. By evaluating six models across four LIBERO suites with varying demonstration counts and training steps, the study finds that sample efficiency varies significantly even among models with similar zero-shot performance. Specifically, the model pi_0.5 requires 2-5 times fewer demonstrations than pi_0, and freezing the vision language model during fine-tuning has a detrimental effect on certain architectures but minimal impact on GR00T models.
Generated from attributed reports · Updated 5 hours ago
Event evidence and corrections
0 attributed source owners. Ownership does not establish independent confirmation. Quantities are reported separately and are never added together.
Reported quantity · demonstrations: 50 other · Basis not reported
“dy adaptation efficiency for six VLAs ($\pi_0$, $\pi_{0.5}$, and GR00T N1-N1.7) on four LIBERO suites, varying both demonstrations and training steps. This reveals differences hidden by full-data fine-tuning: models which are within two points on the full LIBERO-Spatial dataset differ by up to 3$\times$ in success with 50 demonstrations. $\pi_{0.5}$ needs 2-5$\times$ fewer demonstrations than $\pi_0$ and is the most”
Exact source · revision 3Source owner not reported
Reported quantity · training steps: 0 other · Basis not reported
“dy adaptation efficiency for six VLAs ($\pi_0$, $\pi_{0.5}$, and GR00T N1-N1.7) on four LIBERO suites, varying both demonstrations and training steps. This reveals differences hidden by full-data fine-tuning: models which are within two points on the full LIBERO-Spatial dataset differ by up to 3$\times$ in success with 50 demonstrations. $\pi_{0.5}$ needs 2-5$\times$ fewer demonstrations than $\pi_0$ and is the most”
Exact source · revision 3Source owner not reported
Reported quantity · suite count: 4 other · Basis not reported
“dy adaptation efficiency for six VLAs ($\pi_0$, $\pi_{0.5}$, and GR00T N1-N1.7) on four LIBERO suites, varying both demonstrations and training steps. This reveals differences hidden by full-data fine-tuning: models which are within two points on the full LIBERO-Spatial dataset differ by up to 3$\times$ in success with 50 demonstrations. $\pi_{0.5}$ needs 2-5$\times$ fewer demonstrations than $\pi_0$ and is the most”
Exact source · revision 3Source owner not reported
Reported quantity · suites: 3 other · Basis not reported
“sample-efficient model at every budget, with the largest gap on the long-horizon suite. Within GR00T, N1.5 adapts more efficiently than N1.6 and N1.7 on three of four suites, despite their stronger zero-shot performance. We also study one pretraining factor behind these differences by freezing the VLM during finetuning: this severely degrades the $\pi$ models whose VLM is jointly trained with the action expert during”
Exact source · revision 3Source owner not reported
Report timeline
Follow attributed reports and material updates.
- arXiv Robotics — research abstractsSignalHow Many Demonstrations Does a VLA Need? On the Sample Efficiency of VLA Finetuning
This research investigates the data requirements for adapting pretrained vision language action (VLA) models to new tasks. By evaluating six models across four LIBERO suites with varying demonstration counts and training steps, the study finds that sample efficiency varies significantly even among models with similar zero-shot performance. Specifically, the model pi_0.5 requires 2-5 times fewer demonstrations than pi_0, and freezing the vision language model during fine-tuning has a detrimental effect on certain architectures but minimal impact on GR00T models.
Event coverage history
There is not enough continuous observation data to show a trend.
Timezone · UTC
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).