Alibaba QwenChina2026

Qwen-RobotManip

Key figures as stated by the developer

4 of 7 key figures stated

Model type ⁠A vision-language-action
Controls ⁠A arms and hands
Inputs ⁠A camera language robot state
Base model ⁠A Qwen3.5-4B (vision-language backbone)

Not stated: Access · Parameters · Licence

Full specifications

7 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
Released	16 June 2026	[A] "Submitted on 16 Jun 2026" ⁠C
Input and output
Model type	`vision-language-action`	Developer wording, GitHub README: "a generalizable vision-language-action foundation model". ⁠A
Inputs	`camera` `language` `robot state`	README: "The VLM processes multi-view observations, language instructions, structured embodiment prompts, and execution context." Technical report: "The robot’s proprioceptive state is encoded by a two-layer MLP". ⁠A
Controls	`arms and hands`	README: the action space "accommodates single-arm, dual-arm, dexterous hand, and mobile manipulation settings". ⁠A
Degrees of freedom controlled	not stated	Note The count differs per robot; README: "Missing degrees of freedom are zero-padded and excluded from the loss with binary masks". ⁠A
Size and speed
Parameters	not stated	Note The developer names the backbone ("Qwen3.5-4B vision-language backbone") and gives no total. ⁠A
Control rate	not stated
Where it runs and what it is built on
On robot or cloud	not stated
Base model	Qwen3.5-4B (vision-language backbone)	README: "Qwen3.5-4B vision-language backbone". ⁠A
Openness
Access	not stated	Note README: "There is currently no plan to release the model weights for Qwen-RobotManip or Qwen-RobotNav." ⁠A
Licence terms	not stated
Licence	not stated
The developer’s own claims
Learns new tasks	needs retraining	Developer claim, technical report: "we adapt it to specific deployment scenarios through supervised fine-tuning (SFT)". The model also adapts its behaviour in use: README, "allowing the model to adapt without parameter updates". ⁠B
Works across robots	yes	Developer claim, technical report: the model "is validated on real-robot platforms including AgileX ALOHA, Franka, UR, and ARX". ⁠B 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-RobotManip as a generalizable vision-language-action foundation model.
The Qwen-RobotManip technical report was submitted to arXiv on 16 June 2026.

Versions and notes in the key figures (6)

Model type

Developer wording, GitHub README: "a generalizable vision-language-action foundation model".

Controls

README: the action space "accommodates single-arm, dual-arm, dexterous hand, and mobile manipulation settings".

Inputs

README: "The VLM processes multi-view observations, language instructions, structured embodiment prompts, and execution context." Technical report: "The robot’s proprioceptive state is encoded by a two-layer MLP".

Access

Note README: "There is currently no plan to release the model weights for Qwen-RobotManip or Qwen-RobotNav."

Parameters

Note The developer names the backbone ("Qwen3.5-4B vision-language backbone") and gives no total.

Base model

README: "Qwen3.5-4B vision-language backbone".

Sources

A github.com/QwenLM/Qwen-RobotManip · Retrieved 2026-10-19
B arxiv.org/html/2606.17846 · Retrieved 2026-10-19
C arxiv.org/abs/2606.17846 · Retrieved 2026-10-19

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.