2toinf/X-VLA

[ICLR 2026] The offical Implementation of "Soft-Prompted Transformer as Scalable Cross-Embodiment Vision-Language-Action Model"

What it solves

X-VLA addresses the challenge of training generalist Vision-Language-Action (VLA) models that can work across different robotic platforms (embodiments). It leverages heterogeneous robot datasets to create a model that generalizes well across diverse hardware, improving dexterity and adaptability in robotic manipulation tasks.

How it works

X-VLA uses a soft-prompted Transformer architecture. It introduces embodiment-specific learnable embeddings (soft prompts) that guide a unified Transformer backbone to learn policies across multiple domains. To maintain consistency, it employs a unified End-Effector 6D (EE6D) control space for both proprioceptive input and action output.

Who it’s for

This project is for robotics researchers and developers who want to deploy scalable, cross-platform VLA models on physical robots or in simulation platforms like LeRobot.

Highlights

  • Cross-Embodiment Generalization: Achieves state-of-the-art performance across six simulation platforms and three real-world robots.
  • Server-Client Architecture: Separates the model environment from robot-specific dependencies to avoid package conflicts and support distributed inference.
  • Unified Control Space: Uses a standardized EE6D control space to ensure consistency across different robotic arms.
  • Extensible Training: Supports LoRA fine-tuning and a modular dataset interface for adding custom robot demonstrations.

Related

  • Project
    dexmal/opendm

    OpenDM is a Vision-Language-Action (VLA) model framework for open-world robot control, providing tools for training, fine-tuning, and deploying models that can handle complex, multi-embodiment tasks.

  • Dispatch
    Qwen-VLA: Unifying Vision-Language-Action Modeling for Embodied Intelligence

    Qwen-VLA is a general-purpose Vision-Language-Action model that unifies robotic manipulation, vision-language navigation, and cross-embodiment control into a single generalist policy model.

  • Project
    NVIDIA/Isaac-GR00T

    An open vision-language-action (VLA) model for generalized humanoid robot skills, enabling cross-embodiment manipulation tasks via multimodal input.

  • Project
    X-Square-Robot/wall-x

    An open-source framework for embodied foundation models that enables robots to perform complex physical manipulations by bridging vision-language understanding with action prediction.

  • Project
    Open-X-Humanoid/HEX

    A whole-body vision-language-action framework for humanoid robots that enables cross-embodiment manipulation by aligning heterogeneous robot states into shared body-part slots.