Skip to content
RobothecaBeta
Log in

τ0 (tau-0) VLA

  • AI model
  • Humanoids and robot learning

A hierarchical robot foundation model for long-horizon manipulation, pairing a world-model-guided high-level planner with a generalist low-level policy for robot embodiments.

Made by Shanghai Innovation Institute.

Source on GitHub: github.com/sii-research/tau-0-vla (opens in a new tab). Facts read at commit d804047 (opens in a new tab).

Licenses

Code
Apache-2.0 (opens in a new tab)

Read by Robotheca from LICENSE (opens in a new tab) (Apache-2.0) at commit d804047 (opens in a new tab), on 16 Oct 2026.

Commercial use
Yes

What it does

  • A memory-augmented high-level policy generates the next subtask and searches over alternatives with world-model-guided test-time computation when extra reasoning is needed.
  • The low-level policy combines a Qwen3.5 vision-language backbone with a Mixture-of-Transformers action expert trained by conditional flow matching.
  • It uses a unified 40-dimensional state/action space.
  • It was trained on 40,115 hours of heterogeneous real-world robot data with multimodal co-training.
  • Provides post-training code, an example AgiBot World dataset subset, and a LIBERO simulation recipe with checkpoints.
  • Includes serving and open-loop evaluation tools, plus a high-level proposal model and a robotics LoRA world model for goal-image generation.

Written by Robotheca from the project's README (opens in a new tab) at commit d804047.

Getting started

Difficulty
Advanced

Maintenance

Status
Active: a commit in the last 90 days
Newest commit
13 Oct 2026 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.