How Much of Your VLA Is kNN
How Much of Your VLA Is kNN
This research investigates the extent to which Vision-Language-Action (VLA) models rely on generalization versus retrieval. The authors show that retrieving nearby demonstrations achieves a 91.7% success rate on the LIBERO benchmark without policy training, outperforming other methods under perturbations. They introduce Local Action Consistency (LAC), an offline metric to measure action consistency in representation space, which can guide the training of retrieval keys. The study evaluates nine publicly available VLAs and provides source code at https://github.com/sereact/lerobot_knn.
The authors demonstrate that a significant portion of Vision-Language-Action (VLA) model performance on the LIBERO benchmark can be achieved through k-nearest neighbor retrieval without policy training, introducing Local Action Consistency as an offline metric to guide retrieval key training.
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.
- Environment
- SimulationOpen source S2
Reported numbers
Reported success rate
91.7%
Reported trials
View original evidence
arXiv:2611.13160v1 Announce Type: new Abstract: Vision-language-action (VLA) models now exceed 95% success on the LIBERO simulation benchmark, yet it remains unclear how much of this performance requires generalization beyond the training demonstrations. In this work, we show that retrieving nearby demonstrations and reusing their action chunks achieves a success rate of 91.7 % on LIBERO without policy training. Unde
Open source S2
- code: Available Artifact link Open source S4
What remains unknown
Not established in the collected evidence: 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.
arXiv:2611.13160v1 Announce Type: new Abstract: Vision-language-action (VLA) models now exceed 95% success on the LIBERO simulation benchmark, yet it remains unclear how much of this performance requires generalization beyond the training demonstrations. In this work, we show that retrieving nearby demonstrations and reusing their action chunks achieves a success rate of 91.7 % on LIBERO without policy training. Unde
Open source S2
r LIBERO-Plus perturbations, retrieval shows the smallest drop of all evaluated policies. Across nine publicly available VLAs, fine-tuning mostly makes the vision-language backbone a better retrieval key, and the actions of the VLAs remain close to the demonstrations. In addition, we introduce Local Action Consistency (LAC), an offline metric that measures whether nearby observations in representation space are also
Open source S3
associated with similar future actions. We show that LAC tracks the success rate of retrieval and can guide the training of a retrieval key. Our evaluation and source code are available at https://github.com/sereact/lerobot_knn.
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org