RobbyantChina2026

LingBot-VLA

An AI model: a vision-language-action model (VLA) takes camera images and an instruction and turns them into movements for a robot. This page describes the software, not a robot.

RobotsR1 Proby Galaxea Dynamics AG1by Agibot AKUAVO 4Proby Leju Robotics AS1by Astribot F

VersionsLingBot-VLA 1.0LingBot-VLA 2.0

The developer has released this model in several versions. Each figure applies to the version its source names.

Key figures as stated by the developer

5 of 6 stated · 1 gap

Model type ⁠D vision-language-action
Controls ⁠A arms and hands
Inputs ⁠A camera language robot state
Access ⁠D open download
Parameters not stated
Base model ⁠B Qwen3-VL-4B-Instruct

How to read the figures

14kgA30.9lbA: The letter in the grey square after a figure points to the source the figure came from. Robots with several sources carry several letters.
74.5lb: Imperial figures are our own conversion unless the manufacturer states an imperial figure itself. The source letter always belongs to the metric figure — a conversion has no source of its own.

The four states

Versions and notes in the key figures (6)

Model type

Developer wording: "a pragmatic Vision-Language-Action foundation model".

Controls

Version 1.0 was pre-trained on dual-arm robots: "dual-arm robot data collected in the real world for pre-training". Version 2.0 widens this: "the unified representation supports arms, end-effectors, grippers, dexterous hands, waist, head, and mobile-base signals".

Inputs

Paper on version 1.0: the observation has "tokens from three-view operational images", "the task instruction" "and the robot state".

Access

Version 1.0: "Weights and code released!" Version 2.0: "pre-trained weights as a native-depth model."

Parameters

Note The checkpoints are named "LingBot-VLA-4B" (version 1.0) and "lingbot-vla-v2-6b" (version 2.0).

Base model

Each README trains the model from the weights of "Qwen2.5-VL-3B-Instruct" (version 1.0) or "Qwen3-VL-4B-Instruct" (version 2.0).

Versions

LingBot-VLA 1.0: Qwen2.5-VL-3B-Instruct
LingBot-VLA 2.0: Qwen3-VL-4B-Instruct

Full specifications

8 of 12 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
Input and output
Model type	`vision-language-action`	D Developer wording: "a pragmatic Vision-Language-Action foundation model".
Inputs	`camera` `language` `robot state`	A Paper on version 1.0: the observation has "tokens from three-view operational images", "the task instruction" "and the robot state".
Controls	`arms and hands`	A Version 1.0 was pre-trained on dual-arm robots: "dual-arm robot data collected in the real world for pre-training". Version 2.0 widens this: "the unified representation supports arms, end-effectors, grippers, dexterous hands, waist, head, and mobile-base signals".
Degrees of freedom controlled	not stated
Size and speed
Parameters	not stated	Note The checkpoints are named "LingBot-VLA-4B" (version 1.0) and "lingbot-vla-v2-6b" (version 2.0). ⁠C
Control rate	not stated	Note The README gives a latency, not a rate: "one inference call on an NVIDIA GeForce RTX 4090D takes about" 130 ms with 10 denoising steps (version 2.0). ⁠B
Where it runs and what it is built on
Runs where	not stated
Base model	Qwen3-VL-4B-Instruct	B Each README trains the model from the weights of "Qwen2.5-VL-3B-Instruct" (version 1.0) or "Qwen3-VL-4B-Instruct" (version 2.0). Versions LingBot-VLA 1.0 Qwen2.5-VL-3B-Instruct LingBot-VLA 2.0 Qwen3-VL-4B-Instruct
Openness
Access	open download	D Version 1.0: "Weights and code released!" Version 2.0: "pre-trained weights as a native-depth model."
License	Apache-2.0	C Model card of version 1.0: "This project is licensed under the" Apache-2.0 License. The model card of version 2.0 says the same.
The developer’s own claims
Learns new tasks	needs retraining	D Developer claim: the released weights are for post-training ("Pretrained Checkpoints for Post-Training with and without depth") and the paper says "LingBot-VLA can be easily and efficiently transferred to downstream tasks".
Works across robots	yes	E Developer claim: "Through a systematic assessment on 4 robotic platforms" (version 1.0). Version 2.0 shows "strong cross-embodiment long-horizon mobile manipulation capability across the two robotic platforms". 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.

Notes

Robbyant describes LingBot-VLA as a pragmatic vision-language-action foundation model.
Robbyant calls LingBot-VLA 2.0 an advance on LingBot-VLA, the previous version.
The paper on LingBot-VLA 1.0 was submitted to arXiv on 26 January 2026; the paper on 2.0 on 7 July 2026.

Sources

A Manufacturer product page. arxiv.org/html/2601.18692 · Retrieved 2026-10-14
B github.com/Robbyant/lingbot-vla-v2 · Retrieved 2026-10-14
C huggingface.co/robbyant/lingbot-vla-4b · Retrieved 2026-10-14
D github.com/Robbyant/lingbot-vla · Retrieved 2026-10-14
E arxiv.org/abs/2601.18692 · Retrieved 2026-10-14
F arxiv.org/html/2607.06403v1 · Retrieved 2026-10-14

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.