Skip to main content
Rase AI Intelligence Newsletter

Robotics & physical AI Model or AI service

OpenVLA releases a language-guided robot manipulation checkpoint

OpenVLA is a 7-billion-parameter model trained to predict robot actions from images and language instructions. The authors trained it using data from multiple robot setups. [1][2]

Model access
The project’s 7B checkpoint is available through its Hugging Face model card. [1]

What the model does

The authors posted their preprint in June 2024 and made weights available. Their paper describes adapting the model to manipulation tasks on different robot systems. [1][2]

Why it matters

Training on demonstrations from several platforms gives researchers a way to study whether a single policy can be adapted to a robot with different hardware or tasks. [2]

Key dates

  1. OpenVLA preprint first posted [2]

Ways to use it

A checkpoint user needs to align actions and camera observations with the target robot, then verify performance on the intended tasks. [1][2]

Model limitations

The results concern tested robot configurations and tasks. Use of the published checkpoint is subject to the terms for that checkpoint and its base models. [1][2]

Sources and verification

Catalog evidence checked How we verify records

  1. Primary source openvla/openvla-7b
  2. Primary source OpenVLA: An Open-Source Vision-Language-Action Model

Related profiles

Robotics & physical AI

Gemini Robotics 2.5

Google DeepMind

The models are intended to interpret a robot's surroundings and support compatible robot tasks.

High confidenceEvidence checked

Robotics & physical AI

NEURA 4NE1

NEURA Robotics

A humanoid platform NEURA presents for development and industrial training, with multiple distinct hardware generations.

Medium confidenceEvidence checked

Robotics & physical AI

LUMO all-terrain humanoid

Shenzhen Dobot (DOBOT)

LUMO is a humanoid robotics platform whose proposed applications include interaction and service tasks.

Medium confidenceEvidence checked