Skip to content
embodied.guide

RLWRLDUnited States2026

RLDX-1

Key figures as stated by the developer

7 of 7 key figures stated
  • Model type ⁠B vision-language-action
  • Controls ⁠B arms and hands
  • Inputs ⁠A camera language robot state touch
  • Access ⁠A Open weights (download)
  • Parameters ⁠A 6.9B
  • Base model ⁠A Qwen3-VL-8B-Instruct
  • Licence ⁠C Free, research only

Full specifications

12 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
Released5 May 2026

[A] "Submitted on 5 May 2026" ⁠D
Input and output
Model typevision-language-action

RLWRLD's technical report: "RLDX-1 is a Vision-Language-Action model (VLA)". The model card lists the tag "- vision-language-action". ⁠B

Inputscamera language robot state touch

Model card: "RGB video (default 4 frames), state proprioception, optional tactile / torque signals, language instruction." Tactile and torque inputs are optional. ⁠A

Controlsarms and hands

Technical report: "the ability of RLDX-1 to control a high-DoF humanoid robot". The ALLEX test robot: "It is equipped with 7-DoF arms and 15-DoF five-finger hands." ⁠B

Degrees of freedom controllednot stated
Size and speed
Parameters6.9B

Pretrained checkpoint RLDX-1-PT: "Multi-source pretrained foundation (this repo) | 6.9B". The mid-trained checkpoints are larger: "DROID mid-train | 8.1B". ⁠A

Control ratemore than 22Hz

Model inference speed on one GPU, not motor control: "43.7 ms / step on RTX 5090 (1.63× speedup, >22 Hz)". ⁠A
Where it runs and what it is built on
On robot or cloudnot stated
Base modelQwen3-VL-8B-Instruct

Model card header: "base_model: Qwen/Qwen3-VL-8B-Instruct". ⁠A
Openness
AccessOpen weights (download)

Model card: this repository hosts "a foundation checkpoint pretrained on a broad mixture of public manipulation corpora". ⁠A

Licence termsRLWRLD Model License v1.0

License text: the model and derivatives "may only be used for non-commercial purposes". ⁠C

LicenceFree, research only

License text: the model and derivatives "may only be used for non-commercial purposes". ⁠C
The developer’s own claims
Learns new tasksneeds retraining

Developer claim: "For custom robots, finetune." ⁠A

Works across robotsyes

Developer claim: "RLDX-1 covers diverse embodiments including single-arm, dual-arm, and humanoid robots". The model card adds that the pretrained checkpoint "is not guaranteed to work zero-shot on novel embodiments without finetuning". ⁠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

  • RLWRLD publishes a pretrained checkpoint, RLDX-1-PT, from which its task-specific checkpoints are fine-tuned.
  • RLWRLD lists its headquarters in San Francisco, with offices in Seoul and Tokyo, on its own website.
Versions and notes in the key figures (7)

Model type

RLWRLD's technical report: "RLDX-1 is a Vision-Language-Action model (VLA)". The model card lists the tag "- vision-language-action".

Controls

Technical report: "the ability of RLDX-1 to control a high-DoF humanoid robot". The ALLEX test robot: "It is equipped with 7-DoF arms and 15-DoF five-finger hands."

Inputs

Model card: "RGB video (default 4 frames), state proprioception, optional tactile / torque signals, language instruction." Tactile and torque inputs are optional.

Access

Model card: this repository hosts "a foundation checkpoint pretrained on a broad mixture of public manipulation corpora".

Parameters

Pretrained checkpoint RLDX-1-PT: "Multi-source pretrained foundation (this repo) | 6.9B". The mid-trained checkpoints are larger: "DROID mid-train | 8.1B".

Base model

Model card header: "base_model: Qwen/Qwen3-VL-8B-Instruct".

Licence

License text: the model and derivatives "may only be used for non-commercial purposes".

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.