Enhancing Linguistic Generalization of VLA: Fine-Tuning OpenVLA via Synthetic Instruction Augmentation
Dongik Shin · 2026
Read Enhancing Linguistic Generalization of VLA: Fine-Tuning OpenVLA via Synthetic Instruction Augmentation by Dongik Shin (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-16044.txt.
Description
Shin boosts a robot model's language reach with synthetic instruction variety. OpenVLA leads vision-language-action models but falters in fresh scenes. His plan enriches the Bridge Dataset V2 with reworded commands. A large language model writes varied twins of each instruction. The pairs keep meaning while shifting structure and phrasing. LoRA fine-tuning then teaches OpenVLA on the widened set. Few weights move, so training stays cheap and light. Robustness gains suggest language breadth is the key gap. Specialized datasets need richer wording, he argues.
Catalog record
| CREATOR | Dongik Shin |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 16363 |
| LICENSE | http://creativecommons.org/licenses/by-sa/4.0/ |
| RIGHTS | CC BY SA |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.16044