Summary
TurboVLA is a new vision-language-action (VLA) paradigm for robotic control that replaces the conventional LLM-centric V -> L -> A pathway with a direct V + L -> A mapping. Instead of projecting visual observations into a large language model's representation space, TurboVLA independently encodes visual observations and language instructions, exchanges information via lightweight bidirectional vision-language interaction, and predicts continuous action chunks with a compact decoder. This design directly builds task-conditioned representations from vision and language features, substantially cutting the computation and memory cost of each policy invocation. On the LIBERO benchmark, TurboVLA achieves a 97.7% average success rate with only 0.2B parameters, 31.2 ms inference latency, and 0.9 GB of GPU memory on a consumer-grade RTX 4090, matching or exceeding much larger VLA policies. The paper (arXiv:2607.27205) is authored by Hengyi Xie, Chenfei Yao, and colleagues, including teams from HUST-affiliated researchers.
Paper Overview
Field: Computer Vision (CV)
Authors: Hengyi Xie, Chenfei Yao, Xianjin Wu, Xuanyang Xi, Yiping Tang, Di Xu, Yingying Zhu, Dingkang Liang, Xiang Bai, Han Ding
arXiv: 2607.27205
Summary
Vision-language-action (VLA) models commonly adopt an LLM-centric V -> L -> A pathway, where visual observations are projected into the representation space of a large language model before being decoded into robot actions. Although effective, this design incurs substantial computation and memory overhead at every policy invocation.
TurboVLA reformulates the conventional V -> L -> A pathway as a direct V + L -> A mapping. Instead of using a large language model as the central interface between perception and action, TurboVLA:
- Independently encodes visual observations and language instructions
- Exchanges information between them through lightweight bidirectional vision-language interaction
- Predicts continuous action chunks with a compact decoder
This concise design builds task-conditioned representations directly from visual and language features, significantly reducing the computation and memory cost of VLA inference.
Results
On the LIBERO benchmark, TurboVLA achieves a 97.7% average success rate with:
- Only 0.2B parameters
- 31.2 ms inference latency
- 0.9 GB GPU memory
All of this runs on a consumer-grade
RTX 4090, matching or exceeding VLA policies with far more parameters.
---
*Auto-collected on 2026-07-31.*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/178503819