hzxie/DynamicVLA

The official implementation of "DynamicVLA: A Vision-Language-Action Model for Dynamic Object Manipulation". (arXiv 2601.22153)

What it solves

DynamicVLA addresses the challenge of dynamic object manipulation in robotics, allowing robots to interact with objects that are moving or changing state, rather than just static targets.

How it works

It is a Vision-Language-Action (VLA) model that processes visual observations and language instructions to output robotic actions. The project provides a full pipeline for training and evaluation, including a specialized dataset called DOM (Dynamic Object Manipulation) containing 3D scenes and objects. It integrates with Isaac Lab for simulation-based data generation and benchmarking, supporting delta actions (relative movements) and contiguous inference for smoother control.

Who it’s for

Robotics researchers and developers working on embodied intelligence, specifically those focusing on vision-language-action models and the manipulation of dynamic objects.

Highlights

  • DOM Dataset: A dedicated dataset for dynamic object manipulation including 3D scenes and objects.
  • Isaac Lab Integration: Full support for synthetic data generation and benchmarking within a high-fidelity simulation environment.
  • VLA Architecture: Combines vision and language to drive robotic actions.
  • Flexible Action Space: Supports delta actions and various rotation representations (e.g., Euler angles).

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.

  • Project
    InternRobotics/VLAC

    A Vision-Language-Action-Critic model that provides dense reward feedback and trajectory quality screening for real-world robotic reinforcement learning.

  • Project
    Robbyant/lingbot-vla-v2

    LingBot-VLA 2.0 is a Vision-Language-Action foundation model that enables robots to generalize across different embodiments and tasks using a unified action representation and MoE action experts.

  • Project
    2toinf/X-VLA

    X-VLA is a soft-prompted Transformer-based Vision-Language-Action model designed for scalable cross-embodiment robotic control across diverse hardware platforms.

  • Project
    Robbyant/lingbot-vla

    A pragmatic Vision-Language-Action (VLA) foundation model pre-trained on 20,000 hours of real-world robot data to enable high-performance robotic control across multiple platforms.