News Summary

Hugging Face has introduced a new open AI model for robotics named SmolVLA, designed to make building sophisticated robotics projects more accessible. SmolVLA, trained on community-shared datasets, surpasses larger models in performance both virtually and in real-world environments. The model is part of Hugging Face’s initiative to democratize access to vision-language-action models and foster the development of generalist robotic agents. Notably, SmolVLA is compact enough to operate on a single consumer GPU or even a MacBook, and can be deployed on affordable hardware. This model is an extension of Hugging Face’s previous efforts, including the launch of LeRobot and the acquisition of French startup Pollen Robotics. With its asynchronous inference stack, SmolVLA enhances robot response times in dynamic settings.