EmbodiedHub LogoEmbodiedHub
Back to Search
O

OpenVLA

ModelActive

OpenVLA is an open-source VLA model from Stanford and UC Berkeley. Combines Prismatic VLM (SigLIP + Llama 2) with action de-tokenization head. 7B params, trained on Open X-Embodiment. Demonstrated internet-scale VLM pretraining transfers to robot control. Released June 2024. Fully open-source (MIT).

Details

Updated:6/20/2026
open sourcetrue
release date2024-06-13
github urlhttps://github.com/openvla/openvla
paper urlhttps://arxiv.org/abs/2406.09246
model familyVLA (Vision-Language-Action)
huggingface urlhttps://huggingface.co/openvla

Tags

VLAfoundation-modelopen-sourcegeneralistmanipulationStanford7BLLM-for-robotics

Relationships

RoboFlamingo
Model
RELATED_TO
VIMA
Model
RELATED_TO
Language-Table
Dataset
TRAINED_ON
BridgeData V2
Dataset
TRAINED_ON
DROID
Dataset
TRAINED_ON
Open X-Embodiment
Dataset
TRAINED_ON
LeRobot
Project
SUPPORTS
LIBERO
Benchmark
EVALUATED_ON

Sources

https://arxiv.org/abs/2406.09246
website
Visit
https://github.com/openvla/openvla
website
Visit

Appears In

Vision-Language-Action Models

An overview of Vision-Language-Action (VLA) models that enable robots to understand language instructions and perform manipulation tasks.

Embodied AI Landscape

An entry-point knowledge page providing an overview of the embodied AI ecosystem, including companies, robots, models, datasets, and open-source tools.