DreamWAM
- AI model
- Humanoids and robot learning
A research codebase for a world action model that predicts future appearance, motion, depth and semantics to guide robot action prediction, using RGB-only inference.
Made by HUST Vision Lab, China (checked on github.com/hustvl (opens in a new tab)).
Source on GitHub: github.com/hustvl/DreamWAM (opens in a new tab). Facts read at commit 7c35d7d (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 7c35d7d (opens in a new tab), on 16 Oct 2026.
- Commercial use
- Yes
What it does
- Learns future RGB, optical flow, depth and DINO features jointly and uses them to supervise action prediction during training.
- Beyond-RGB signals are used only in training, so inference needs RGB input alone.
- Built on the FastWAM codebase and a shared VideoDiT-ActionDiT attention design.
- Provides released checkpoints on Hugging Face and training configs for joint and unconditional settings.
- Includes evaluation scripts for the four LIBERO suites and the LIBERO-Plus protocol, writing JSON results.
- Requires preparing several upstream components (Wan2.2, RAFT, DINOv2, Depth Anything 3, LIBERO) and a precomputed training cache.
Written by Robotheca from the project's README (opens in a new tab) at commit 7c35d7d.
Getting started
- Difficulty
- Advanced
- How to start
- The project's README (opens in a new tab)
Maintenance
- Status
- Active: a commit in the last 90 days
- Newest commit
- 28 Aug 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.