OpenVLA: the first open-source Vision-Language-Action model for generalist robotics
In one sentence Berkeley and Stanford researchers release OpenVLA, 7B parameters, the first open-source VLA for generalist robot control — a universal controller downloadable from Hugging Face.
OpenVLA is an AI model you can freely download and use to control a robot. Before OpenVLA, models of this type were all proprietary and accessible only to large companies.
With 7 billion parameters, OpenVLA takes camera images and natural language instructions as input, and produces movements to execute. It's trained on data from over 970,000 robot episodes from various sources.
Being open source means any university lab or independent researcher can now build on this model, enormously accelerating robotics research.
Companies
UC Berkeley, Stanford, Google DeepMind
Tools
OpenVLA, Prismatic VLM
Tags
Sources