Skip to content
RobothecaBeta
Log in

RynnWorld-4D

  • AI model
  • Humanoids and robot learning

RynnWorld-4D is an open 4D embodied world model that jointly generates RGB, depth, and optical flow to predict scene evolution for robotic manipulation, plus a policy head for bimanual control.

Made by Alibaba DAMO Academy.

Source on GitHub: github.com/alibaba-damo-academy/RynnWorld-4D (opens in a new tab). Facts read at commit 2c74853 (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 2c74853 (opens in a new tab), on 16 Oct 2026.

Commercial use
Yes

What it does

  • Co-generates synchronized RGB, depth, and optical flow so predictions carry 3D geometry and motion, not just pixels.
  • Uses a tri-branch diffusion transformer with mutual cross-modal attention across appearance, geometry, and motion.
  • Trains in three stages: SFT warm-up, tri-branch RoPE positional encoding, then joint-attention fine-tuning with decaying depth/flow-to-RGB injection.
  • Provides inference that outputs an rgb.mp4, depth.mp4, and flow.mp4 per sample.
  • RynnWorld-4D-Policy is a lightweight flow-matching action head on a frozen world-model backbone, serving actions over a websocket protocol.
  • Ships sample data for smoke-testing training and the policy pipeline.

Written by Robotheca from the project's README (opens in a new tab) at commit 2c74853.

Getting started

Difficulty
Advanced

Maintenance

Status
Maintained: a commit in the last 12 months
Newest commit
16 Jul 2026 on main
Latest release
None published on GitHub

Checked on GitHub on 16 Oct 2026.

Links

Robots that can run 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.