TurboVLA: Real-Time Vision-Language-Action Model at 32 Hz on an RTX 4090 with <1 GB VRAM

2026-07-29

AI Summary

This work proposes TurboVLA, an efficient vision-language-action (VLA) framework that eliminates the reliance on large language models (LLMs) as a central hub between perception and action, thereby significantly reducing computational overhead and GPU memory consumption. Instead of LLM-centric architectures, TurboVLA establishes an end-to-end V+L→A mapping by independently encoding visual and linguistic inputs, fusing them through a lightweight bidirectional interaction module, and predicting continuous actions via a compact decoder. With only 0.2 billion parameters, TurboVLA achieves an average success rate of 97.7% on the LIBERO benchmark, while maintaining a low inference latency of 31.2 ms and consuming merely 0.9 GB of GPU memory. On an RTX 4090, it delivers real-time performance at 32 Hz, matching or surpassing substantially larger models in both efficiency and effectiveness.

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.

Problem

• vision-language-action
• computational overhead
• memory efficiency
• real-time inference
• robotic manipulation

Innovation

• Vision-Language-Action
• Efficient Robotics
• Direct V+L to A Mapping
• Lightweight Interaction
• Real-Time Inference