Skip to content
embodied.guide

Alibaba QwenChina2026

Qwen-RobotWorld

Key figures as stated by the developer

2 of 7 key figures stated
  • Inputs ⁠A camera language
  • Base model ⁠A Qwen2.5-VL (frozen, as action encoder)

Not stated: Model type · Controls · Access · Parameters · Licence

Full specifications

3 of 14 fields stated
The letter points to the source at the foot of the page. Fields without a figure are dashed: the developer does not state them — that is not the same as zero, and not the same as no.
Field Value Source and caveats
Release
Released15 June 2026

[A] "Submitted on 15 Jun 2026" ⁠A
Input and output
Model typenot stated

Note The developer calls it a video world model and does not label it vision-language-action, world action model or embodied reasoning: "We introduce Qwen-RobotWorld, a language-conditioned video world model for embodied intelligence." ⁠A

Inputscamera language

Abstract: "it predicts physically grounded future visual trajectories from current observations across robotic manipulation, autonomous driving, indoor navigation, and human-to-robot transfer"; "With natural language as a unified action interface". Technical report: "states are visual observations (video frames or their latent representations)". ⁠A

Controlsnot stated

Note The model outputs video, not robot commands: "it predicts physically grounded future visual trajectories from current observations across robotic manipulation, autonomous driving, indoor navigation, and human-to-robot transfer". ⁠A

Degrees of freedom controllednot stated
Size and speed
Parametersnot stated

Note The developer gives parts, not a total: "Total parameters: MLLM 7B, VAE 127M (encoder 54M + decoder 73M), MMDiT 20B." ⁠B

Control ratenot stated
Where it runs and what it is built on
On robot or cloudnot stated
Base modelQwen2.5-VL (frozen, as action encoder)

Abstract: a 60-layer double-stream diffusion transformer that "couples frozen Qwen2.5-VL semantics with video-VAE latents". ⁠A
Openness
Accessnot stated
Licence termsnot stated
Licencenot stated
The developer’s own claims
Learns new tasksnot stated
Works across robotsnot stated

Yes only if the developer itself says the model runs on different robot models, for example robots from two manufacturers or two of one manufacturer’s models. Several units of the same robot do not count.
How to read the figures
50HzA
The letter in the grey square after a figure points to the source the figure came from. Pages with several sources carry several letters.

What the data labels mean

Notes

  • The Qwen team describes Qwen-RobotWorld as a language-conditioned video world model for embodied intelligence.
  • The Qwen-RobotWorld technical report was submitted to arXiv on 15 June 2026.
Versions and notes in the key figures (5)

Model type

Note The developer calls it a video world model and does not label it vision-language-action, world action model or embodied reasoning: "We introduce Qwen-RobotWorld, a language-conditioned video world model for embodied intelligence."

Controls

Note The model outputs video, not robot commands: "it predicts physically grounded future visual trajectories from current observations across robotic manipulation, autonomous driving, indoor navigation, and human-to-robot transfer".

Inputs

Abstract: "it predicts physically grounded future visual trajectories from current observations across robotic manipulation, autonomous driving, indoor navigation, and human-to-robot transfer"; "With natural language as a unified action interface". Technical report: "states are visual observations (video frames or their latent representations)".

Parameters

Note The developer gives parts, not a total: "Total parameters: MLLM 7B, VAE 127M (encoder 54M + decoder 73M), MMDiT 20B."

Base model

Abstract: a 60-layer double-stream diffusion transformer that "couples frozen Qwen2.5-VL semantics with video-VAE latents".

Sources

One line per source, not per figure: a source has one retrieval date. The letter beside a figure points to the line it came from.