OPEN MODELS/VISION-LANGUAGE-ACTION/OPENVLA
ADVANCEDVISION-LANGUAGE-ACTIONFEATURED

OpenVLA

State-of-the-art open-source VLA for robot manipulation

WHAT DOES THIS ACTUALLY DO?

This is one of the first open-source models that lets a robot understand what you say — like 'pick up the red cup' — and actually do it. Think of it as giving a robot something close to common sense for physical tasks.

PARAMETERS
7B
RELEASED
2024-06
LICENSE
MIT
ARCHITECTURE
SigLIP...
CAN I ACTUALLY RUN THIS?
6
Intermediate
OUT OF 10

A serious ML setup is needed — 24GB GPU and Python environment. Researchers and ML engineers will be fine. Hobbyists may struggle without cloud access.

DOWNLOAD WEIGHTS
Download from Hugging FaceView on GitHub
15GB disk space required
OVERVIEW

A 7-billion parameter Vision-Language-Action model trained on 970K robot episodes from Open X-Embodiment. Controls multiple robots out-of-the-box and supports efficient fine-tuning.

TRAINED ON

Open X-Embodiment (970K episodes across 22 robots)

ARCHITECTURE
SigLIP + DinoV2 → Projector → Llama 2 7B
CAPABILITIES
Multi-robot control
Language-conditioned manipulation
Goal-image conditioning
Parameter-efficient fine-tuning (LoRA)
Real-time inference
REAL ROBOTICS USE CASES
WAREHOUSE

Sort packages by spoken category

RESEARCH LAB

Baseline for manipulation experiments

MANUFACTURING

Assembly step guidance via language

HARDWARE REQUIREMENTS
MIN VRAM
16GB
REC VRAM
24GB
MIN RAM
32GB
DISK SPACE
15GB
JETSON
Not Supported
CPU ONLY
GPU Required
TESTED GPUs
RTX 3090RTX 4090A100H100
DEPLOYMENT PATHS
Local PC
Requires 24GB VRAM or quantizationSUPPORTED
Docker
Official image availableSUPPORTED
ROS2
Community ROS2 wrappers availableSUPPORTED
Jetson
Too large for current Jetson hardwareNOT SUPPORTED
Cloud (A100)
Recommended for productionSUPPORTED
Isaac Sim
NOT SUPPORTED
RELATED PAPERS
Open X-Embodiment: Robotic Learning Datasets and RT-X Models
RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control
READ ON ARXIV →
RELATED ORGANIZATIONS
Google DeepMindToyota Research InstitutePhysical Intelligence
ORGANIZATIONS
StanfordUC BerkeleyToyota ResearchGoogle DeepMind
DIFFICULTY
ADVANCED

Serious GPU and ML engineering required

COMPUTE2/10
SETUP4/10
BEGINNER-FRIENDLY4/10
EDGE DEPLOYMENT
Jetson Not Supported
CPU-only Not Available
Community →

Seeded and planned prompts — not live forum activity yet.