VICTR: Value-Informed In-Context Learning for Vision-Language-Action Models
VICTR: Value-Informed In-Context Learning for Vision-Language-Action Models
This research presents VICTR, a framework enabling in-context learning in vision-language-action models through progress-informed context retrieval. By estimating task-relative progress using an In-Context Value Function Estimator conditioned on reference demonstrations, the system selects relevant demonstration chunks to guide action generation. On unseen out-of-domain real-world manipulation tasks, VICTR improved the success rate from 6.2% to 23.8% and increased the task progress rate from 28.9% to 55.5%, demonstrating significant gains over a fine-tuned baseline without further fine-tuning.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
Reported success rate
6.2%
Reported trials
View original evidence
d for in-context action generation. On unseen out-of-domain real-world manipulation tasks, VICTR raises the success rate from 6.2% for a fine-tuned $\pi_{0.5}$ baseline to 23.8% and more than doubles the task progress rate, from 28.9% to 55.5%. Project page with code, data and videos: https://ai4ce.github.io/VICTR/
Open source S5Reported success rate
23.8%
Reported trials
View original evidence
d for in-context action generation. On unseen out-of-domain real-world manipulation tasks, VICTR raises the success rate from 6.2% for a fine-tuned $\pi_{0.5}$ baseline to 23.8% and more than doubles the task progress rate, from 28.9% to 55.5%. Project page with code, data and videos: https://ai4ce.github.io/VICTR/
Open source S5
- paper: not_reported
- code: Reported pending Open source S5
- weights: not_reported
- dataset: not_reported
What remains unknown
Not established in the collected evidence: Environment, Control, Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
d for in-context action generation. On unseen out-of-domain real-world manipulation tasks, VICTR raises the success rate from 6.2% for a fine-tuned $\pi_{0.5}$ baseline to 23.8% and more than doubles the task progress rate, from 28.9% to 55.5%. Project page with code, data and videos: https://ai4ce.github.io/VICTR/
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org