NVIDIA · robot AI model
Long-WAM
United States · 2026
Runs on G1 by Unitree RoboticsSource and note: G1B arxiv.org/abs/2610.10528 · Retrieved 19 November 2026
Sources retrieved: 19 November 2026 (A, B, C)
- Model type
- World action modelSource and note: Model typeB arxiv.org/abs/2610.10528 · Retrieved 19 November 2026Note. “We present Long-WAM, a model-system framework for scaling the context of causal world-action models under real-time control constraints.”
- What it controls
- Arms and handsSource and note: What it controlsB arxiv.org/abs/2610.10528 · Retrieved 19 November 2026Note. “Real-time deployment on Unitree G1 and YAM supports dynamic and long-horizon manipulation”. “Generate two future-video latents using 4 steps (imagination sigma 0.9), then predict a 32-step action chunk using 10 action steps.”
- Inputs
- CameraSource and note: InputsA nvlabs.github.io/LongLive/Long-WAM/ · Retrieved 19 November 2026Note. Processes “Observed video history”: “P48 context uses 48 past control steps plus the current observation: **2.4 seconds at 20 Hz**, encoded into four clean latent frames.”
- Access
- Open weights (download)Source and note: AccessC huggingface.co/Efficient-Large-Model/Long-WAM-RoboCasa365 · Retrieved 19 November 2026Note. “Long-WAM policy weights for RoboCasa365.”
- Base model
- LongLive2.0-RobotSource and note: Base modelA nvlabs.github.io/LongLive/Long-WAM/ · Retrieved 19 November 2026Note. Video expert is “initialized from LongLive2.0-Robot”.

Specifications
8 of 14 stated 6 not stated
4 fields hidden 2 not-stated fields with a note stay ·
How to read the figuresShow guideClose
- not statedNo source
- yesStated by a source
- 0Reported as zero
- ASource
Select a source marker for source details and notes.
Release1/1 disclosed
Input and output3/4 disclosed
Size and speed1/2 disclosed
Where it runs and what it is built on1/2 disclosed
Openness1/3 disclosed
The developer’s own claims1/2 disclosed
Notes
Developed by researchers from NVIDIA, MIT, HKU, and UCSD.
Described by the authors as “We present Long-WAM, a model-system framework for scaling the context of causal world-action models under real-time control constraints.”
Sources
One line per source, not per figure: a source has one retrieval date.
- Anvlabs.github.io/LongLive/Long-WAM/Main source. Retrieved 19 November 2026
- Barxiv.org/abs/2610.10528Retrieved 19 November 2026
- Chuggingface.co/Efficient-Large-Model/Long-WAM-RoboCasa365Retrieved 19 November 2026
More models
Same developer

Cosmos 3NVIDIA

Cosmos PolicyNVIDIA

Cosmos Reason 2NVIDIA
- No photograph
DreamDojoNVIDIA

DreamZeroNVIDIA
- No photograph
GR00T 2NVIDIA



