Skip to content
RobothecaBeta
Log in

MiniVLA

  • AI model
  • Humanoids and robot learning

MiniVLA is a smaller variant of the OpenVLA vision-language-action model, built on a Qwen2.5 0.5B backbone, for training robot manipulation policies from images and language instructions.

  • Stale

Made by Stanford Intelligent and Interactive Autonomous Systems Group (ILIAD).

Source on GitHub: github.com/Stanford-ILIAD/openvla-mini (opens in a new tab). Facts read at commit 0822b36 (opens in a new tab).

Licenses

Code
MIT (opens in a new tab)

Read by Robotheca from LICENSE (opens in a new tab) (MIT) at commit 0822b36 (opens in a new tab), on 16 Oct 2026.

Commercial use
Yes

Labels

  • StaleNo commit in 12 months: the newest is from 19 Mar 2025.

What it does

  • Adds MiniVLA configs using a Qwen2.5 0.5B backbone, with support for training on LIBERO-90 data.
  • Supports multi-image input, including wrist images and image histories.
  • Adds residual VQ action chunking, which compresses action sequences into discrete code words.
  • Includes a simpler evaluation option similar to LIBERO, in experiments/robot/simpler/run_simpler_eval.py.
  • Provides a base Prismatic VLM trained on the Qwen2.5 0.5B backbone, plus pretrained LIBERO checkpoints on Hugging Face.
  • Converting MiniVLA models to Hugging Face format and deploying VQ or multi-image variants is not yet supported.

Written by Robotheca from the project's README (opens in a new tab) at commit 0822b36.

Getting started

Difficulty
Advanced

Maintenance

Status
Stale: no commit in 12 months
Newest commit
19 Mar 2025 on main
Latest release
None published on GitHub

Checked on GitHub on 16 Oct 2026.

Links

Robots on Robotheca that use it

No robot on Robotheca is linked to it yet. Links come from a robot's robot.yaml, from its dependency files read as text, or from Robotheca's editors.

Robotheca lists AI models as a reference: nothing is hosted or run here, and every fact links to where it was read.