Skip to content
RobothecaBeta
Log in

DreamZero

  • AI model
  • Humanoids and robot learning

DreamZero is a research project from NVIDIA's GEAR Lab: a World Action Model that jointly predicts robot actions and video, aimed at zero-shot performance on unseen tasks.

  • Non-commercial weights

Made by NVIDIA GEAR Lab.

Source on GitHub: github.com/dreamzero0/dreamzero (opens in a new tab). Facts read at commit ab790c1 (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 ab790c1 (opens in a new tab), on 16 Oct 2026.

Commercial use
No

The weights license does not allow commercial use.

Labels

  • Non-commercial weightsThe weights license (CC-BY-NC-4.0) does not allow commercial use.

What it does

  • Jointly predicts actions and videos, targeting strong zero-shot performance on tasks it has not been trained on.
  • Provides a pretrained DreamZero-DROID checkpoint (14B) for inference, plus a DreamZero-AgiBot checkpoint for post-training on new embodiments.
  • Includes a distributed WebSocket inference server with DiT caching, reported at about 0.6s per inference on GB200 and about 3s on H100.
  • Supports video generation and saving as MP4, and simulation evaluation via sim_evals and RoboArena integration.
  • Ships LoRA and full fine-tuning training scripts, plus the preprocessed DROID dataset and a guide for adding new embodiments.
  • Built on Wan2.1-I2V-14B-480P with the umt5-xxl tokenizer.

Written by Robotheca from the project's README (opens in a new tab) at commit ab790c1.

Getting started

Difficulty
Advanced
Hardware it needs, as the project states it
Python 3.11 and a multi-GPU setup with CUDA 12.9+ (tested on GB200 and H100), minimum 2 GPUs for distributed inference; flash attention required, Transformer Engine and TensorRT only for GB200.

Maintenance

Status
Maintained: a commit in the last 12 months
Newest commit
19 Apr 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.