Paper 2607.27205
TurboVLA: Real-Time Vision-Language-Action Model at 32 Hz on an RTX 4090 with <1 GB VRAM
- Published
- Jul 2026
- Research lab
- Independent
- Citations
- 0
- GitHub
- 385 stars
01 In brief
Summary
TurboVLA introduces a new paradigm for vision-language-action (VLA) models, replacing the conventional LLM-centric V→L→A pathway with a direct V+L→A mapping.
It uses a lightweight BERT text encoder and a DINOv3 vision encoder, with bidirectional cross-attention to fuse visual and linguistic features, and an ACT-style decoder to predict continuous action chunks in a single forward pass.
This design avoids the computational and memory overhead of large language models.
On LIBERO, TurboVLA achieves 97.7% average success with only 0.2B parameters, 31.2 ms latency, and 0.9 GB VRAM on an RTX 4090, outperforming larger models like π0.5 (96.9%) and VLA-JEPA (97.2%).
It also excels on RoboTwin 2.0 bimanual tasks (60.2% success, 43.4 ms) and in real-world tests on an AgileX Piper platform, consistently beating π0.5.
Ablations confirm the importance of semantic language conditioning, bidirectional interaction, and optimal interaction depth (N=6) and action horizon (H=12).
TurboVLA demonstrates that execution-level control does not require an LLM core, offering a more efficient and deployable alternative for robotic manipulation.
Future work will explore hierarchical systems combining LLM planning with TurboVLA's efficient execution.
02 From the paper
Abstract
Vision-language-action (VLA) models commonly adopt an LLM-centric $V \to L \to A$ 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. In this work, we introduce TurboVLA, a new VLA paradigm that reformulates the conventional $V \to L \to A$ pathway as a direct $V + L \to A$ mapping. Instead of using a large language model as the central interface between perception and action, TurboVLA independently encodes visual observations and language instructions, directly exchanges information between them through lightweight bidirectional vision-language interaction, and predicts continuous action chunks with a compact decoder. This simple design constructs task-conditioned representations directly from visual and linguistic features, significantly reducing the computational and memory costs of VLA inference. On LIBERO, TurboVLA achieves 97.7% average success with only 0.2B parameters, 31.2 ms inference latency, and 0.9 GB inference VRAM on a consumer-grade RTX 4090, matching or outperforming substantially larger VLA policies. These results establish TurboVLA as a simple and effective alternative to the prevailing LLM-centric VLA paradigm, offering a new perspective on how vision, language, and action can be connected for efficient robotic manipulation. Code is available at https://github.com/H-EmbodVis/TurboVLA.