Building advanced robotics projects at home is gradually getting simpler. This week, Hugging Face, an AI development platform, introduced SmolVLA, an open-source robotics model.
According to the company, SmolVLA—trained on community-contributed datasets with compatible licenses—surpasses much larger robotics models in both simulated and real-world settings.
In a blog post, Hugging Face states that “SmolVLA is designed to make vision-language-action (VLA) models more accessible and to drive progress in general-purpose robotics.” The company describes it not only as a compact yet powerful model but also as a framework for training and evaluating generalist robotics systems.
Building an Affordable Robotics Ecosystem
SmolVLA is part of Hugging Face’s growing initiative to build a low-cost robotics ecosystem. Last year, it introduced LeRobot, a suite of robotics-specific models, datasets, and tools. More recently, the company acquired French startup Pollen Robotics and launched several affordable robotics platforms, including humanoid robots.
Hugging Face trained SmolVLA, a model with 450 million parameters, using data from the LeRobot Community Datasets—robotics-specific datasets shared on its AI development platform. Parameters, also known as “weights,” are the internal elements of a model that influence how it functions.
Hugging Face says SmolVLA runs on a single consumer GPU—or even a MacBook—and lets users test and deploy it on low-cost hardware, including the company’s own robotics platforms.
Faster Robot Responses Through Asynchronous Processing
In a notable feature, SmolVLA includes an “asynchronous inference stack,” which, according to Hugging Face, enables the model to handle a robot’s actions separately from its sensory input—what it sees and hears. This design, the company explains in a blog post, allows robots to react more swiftly in dynamic, rapidly changing environments.
It’s important to point out that Hugging Face isn’t the only contender in the emerging open robotics space.
Nvidia offers its own suite of open robotics tools, while startup K-Scale Labs is developing components for what it describes as “open-source humanoids.” Other notable players in the field include Dyna Robotics, Physical Intelligence (backed by Jeff Bezos), and RLWRLD.
Read the original article on: TechCrunch
Read more: New System Improves Machine Recognition of Facial Expressions
