Hugging Face claims its new robotics model can run on a MacBook.

It’s getting easier to build robotics projects.

Earlier in the week, Hugging Face’s AI dev platform released an open AI robotics model called SmolvlaHugging Face claims that SmolVLA, which is trained on “compatibly-licensed” community-shared datasets outperforms larger models in both virtual and physical environments.

“SmolVLA aims to democratize access to vision-language-action [VLA] models and accelerate research toward generalist robotic agents,” writes Hugging Face in a blog post. “SmolVLA not only is a lightweight but capable model, it is also a method of training and evaluating generalist robots [technologies].”

SmolVLA forms part of Hugging Face’s rapidly expanding effort in establishing an ecosystem of low cost robotics hardware andsoftware. LeRobot is a collection that includes robotics-focused datasets, models, and tools. Hugging Face recently acquired Pollen Robotics – a robotics start-up based in France – and released several affordable robotics systems for sale, including humanoids.

SmolVLA was trained using data from LeRobot Community Datasets. These datasets are specially marked robotics datasets that are shared on Hugging Face AI development platform. The internal components that determine a model’s behavior are called parameters, or “weights”.

Hugging Face claims SmolVLA can run on a single GPU consumer — or even on a MacBook — it can be tested on “affordable hardware”including the company’s robotics systems.

SmolVLA supports an “asynchronous stack of inferences” which, according to Hugging Face, allows the model’s processing of robot actions from what it hears and sees. The company explains that “[b]Because of this separation, robots are able to respond more quickly in rapidly changing environments.”

SmolVLA can be downloaded from Hugging Face. A user on X has claimed to have used this model to control a robotic arm from a third party:

SmolVLA – feels like a BERT for robotics.
On the Koch Arm, I tried it:
Fine-tuned on RTX2050 (4GB), and matches/outperforms baselines on single-task

Thanks to @RemiCadene @danaubakirova @ Mustash97 @francesco__capu””https://t.co/TiBkAZGwkM”” rel=””nofollow” “> pic.twitter.com/TiBkAZGwkM

— Xingdong Zuo (@XingdongZ)””https://twitter.com/XingdongZ/status/1930168794099659147?ref_src=twsrc%5Etfw”” rel=””nofollow” “> June 4, 2025

Hugging Face is not the only player in this nascent race of open robotics.

Nvidia offers a set of tools for open robots, while K-Scale Labs builds the components for “open-source humans.” Other notable firms in this segment include Dyna Robotics and Jeff Bezos-backed Physical Intelligence.

Kyle Wiggers, TechCrunch AI Editor. His writings have appeared in VentureBeat, Digital Trends and a variety of gadget blogs, including Android Police and Android Authority, Droid-Life and XDA-Developers. He lives in Manhattan, with his music therapist partner.

View Bio

www.aiobserver.co

More from this stream

Recomended