SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics
This work addresses the problem of making robotics more accessible and affordable for researchers and practitioners by reducing training and inference costs, though it is incremental in optimizing existing VLA approaches.
The paper tackles the high computational cost and limited deployability of large vision-language-action (VLA) models for robotics by introducing SmolVLA, a small and efficient model that achieves performance comparable to VLAs 10 times larger while being trainable on a single GPU and deployable on consumer-grade hardware.
Vision-language models (VLMs) pretrained on large-scale multimodal datasets encode rich visual and linguistic knowledge, making them a strong foundation for robotics. Rather than training robotic policies from scratch, recent approaches adapt VLMs into vision-language-action (VLA) models that enable natural language-driven perception and control. However, existing VLAs are typically massive--often with billions of parameters--leading to high training costs and limited real-world deployability. Moreover, they rely on academic and industrial datasets, overlooking the growing availability of community-collected data from affordable robotic platforms. In this work, we present SmolVLA, a small, efficient, and community-driven VLA that drastically reduces both training and inference costs, while retaining competitive performance. SmolVLA is designed to be trained on a single GPU and deployed on consumer-grade GPUs or even CPUs. To further improve responsiveness, we introduce an asynchronous inference stack decoupling perception and action prediction from action execution, allowing higher control rates with chunked action generation. Despite its compact size, SmolVLA achieves performance comparable to VLAs that are 10x larger. We evaluate SmolVLA on a range of both simulated as well as real-world robotic benchmarks and release all code, pretrained models, and training data.