Hugging Face, the global AI model-sharing platform, has unveiled a groundbreaking open-source AI model for robotics called SmolVLA, as reported by TechCrunch.
In a blog post, Hugging Face asserted that SmolVLA, trained on community-shared datasets, outshines large-scale robot AI models and exhibits exceptional performance in both virtual and real-world scenarios.
The company described SmolVLA as a lightweight yet powerful model designed to improve access to vision-language-action (VLA) models and accelerate research into general-purpose robotic agents. They added that it’s not just a model, but also a methodology for training and evaluating universal robotic technologies.
With 450 million parameters, SmolVLA was reportedly trained using robotics datasets available on Hugging Face’s AI development platform. The company emphasized that this model is versatile enough to run on consumer-grade GPUs or MacBooks and can be tested and deployed on cost-effective hardware, including Hugging Face’s robotics systems.
A standout feature of SmolVLA is its support for an asynchronous inference stack. According to Hugging Face, this capability enables robots to process visual input and actions independently, allowing for rapid responses in dynamic environments.
SmolVLA is now available for download on the Hugging Face platform. Notably, a user on X has already claimed success in using the model to control a third-party robotic arm.