Hugging Face Launches Lightweight SmolVLA Model, Enabling Advanced Robotics on Consumer Devices Like MacBooks

Hugging Face has announced the release of SmolVLA, a new robotics model designed to make advanced robotics projects more accessible. Launched earlier this week, SmolVLA boasts impressive efficiency and performance, capable of running on consumer hardware like a MacBook. According to Hugging Face, this model outperforms much larger counterparts in both virtual and real-world environments. In a recent blog post, Hugging Face states that SmolVLA aims to democratize access to vision-language-action (VLA) models, accelerating research into generalist robotic agents. The company emphasizes that SmolVLA is not just a lightweight, high-capability model but also a robust method for training and evaluating robotics technologies. This development is part of Hugging Face's broader initiative to create a low-cost robotics ecosystem. In the past year, the company has made significant strides in this direction with the launch of LeRobot, a suite of robotics-focused models, datasets, and tools. Additionally, Hugging Face acquired Pollen Robotics, a French robotics startup, and introduced several affordable robotics systems, including humanoid models, for purchase. SmolVLA, with 450 million parameters, was trained using data from LeRobot Community Datasets, a set of specially marked robotics datasets shared on Hugging Face's platform. These parameters, or weights, are the internal components that dictate the model's behavior. Hugging Face claims that SmolVLA's compact size allows it to run seamlessly on a single consumer GPU, or even a MacBook, making it accessible to a wider range of developers and enthusiasts. One of the distinctive features of SmolVLA is its support for an "asynchronous inference stack." This feature enables the model to process a robot's actions independently from what it perceives through its sensors. By separating these processes, robots can react more swiftly in dynamic environments, enhancing their practicality and efficiency. The model is now available for download on Hugging Face's platform. Enthusiasts and developers have already begun experimenting with it. One user on X (formerly Twitter) reported using SmolVLA to control a third-party robotic arm, demonstrating its versatility and ease of integration. While Hugging Face is a notable player in the open robotics movement, it is not alone. Competitors such as Nvidia offer a range of tools for open robotics, and startups like K-Scale Labs are developing components for open-source humanoids. Other significant firms in this field include Dyna Robotics, Jeff Bezos-backed Physical Intelligence, and RLWRLD. These companies are all striving to make robotics more accessible and affordable, fostering a vibrant and competitive market. The release of SmolVLA signifies a significant step forward in Hugging Face's mission to empower researchers and developers with powerful yet cost-effective robotics solutions. By lowering the barriers to entry, Hugging Face hopes to spur innovation and collaboration in the robotics community, ultimately advancing the field of generalist robotic agents.