OpenVLA authors · Research paper

OpenVLA: An Open-Source Vision-Language-Action Model

The authors’ project page connects model design, demonstration data, fine-tuning and evaluation resources.

Source text reviewed October 27, 2026. Study notes and exercises are written by botstest.

Read the original ↗

Where to start

Read the model and data descriptions, then follow the paper/code for the evaluation you intend to reproduce.

Use it in your project

Map the model observations and actions to a chosen task before comparing checkpoints.

Scope and limitations

Reported experiments have specific embodiments and setups. They do not establish unrestricted autonomy for an arbitrary robot.

Follow a related roadmap

Evaluate vision-language-action models responsibly · Beginner →

Read: Read the model and data descriptions, then follow the paper/code for the evaluation you intend to reproduce.

Use the model and embodiment description to decide which evaluation assumptions your setup can actually support.

Produce: Map the model inputs, outputs and embodiment assumptions to one candidate evaluation task.