---
title: DM0.5 Vision-Language-Action
emoji: 🤖
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 6.20.0
app_file: app.py
short_description: DM0.5 VLA — predict robot actions from images + text
python_version: "3.12"
startup_duration_timeout: 1h
---

# DM0.5 — Vision-Language-Action robot control

Interactive demo of [**Dexmal/DM05**](https://huggingface.co/Dexmal/DM05), an open-world
Vision-Language-Action (VLA) foundation model (Gemma3-4B VLM backbone + 680M action expert).

Given three robot camera views (head + left/right wrist) and a natural-language instruction,
the model predicts a chunk of 50 future bimanual actions (6 joint targets + 1 gripper command
per arm) using flow-matching, which the demo renders as per-joint / per-gripper trajectory plots.

Built with [OpenDM](https://github.com/dexmal/opendm). Robot type: DOS W1 (14-dim state).