dexmal/dexbotic

Dexbotic: Open-Source Vision-Language-Action Toolbox

What it solves

Dexbotic provides a unified development toolbox for Vision-Language-Action (VLA) models, simplifying the process of pretraining, fine-tuning, inference, and evaluation for embodied intelligence. It removes the friction of setting up complex environments for different VLA algorithms and provides a standardized way to handle training data across various robot platforms.

How it works

Built on PyTorch, the toolbox uses a modular architecture consisting of layered configurations, factory registration, and entry dispatch. This allows researchers to swap models, modify configurations, or add new tasks via experimental scripts without rewriting core code. It supports both local consumer-grade GPUs and cloud training platforms, and includes a unified training data format to ensure compatibility across different hardware like UR5, Franka, and ALOHA.

Who it’s for

It is designed for researchers and developers working on embodied AI, specifically those developing VLA models for robotic manipulation and navigation.

Highlights

  • Broad Model Support: Integrates mainstream policies including π0, CogACT, OFT, MemVLA, and DM0.
  • Optimized Performance: Includes high-performance pre-trained foundation models and a Triton-backed realtime backend that can provide up to a 5x inference speedup.
  • Flexible Training: Supports FSDP2 for distributed training and LoRA for efficient fine-tuning.
  • Hardware Agnostic: Provides deployment scripts and data formats for multiple robot types and supports the latest Blackwell GPU architecture.

Related

  • Project
    OpenBMB/DeepThinkVLA

    DeepThinkVLA is a Vision‑Language‑Action model that inserts an explicit chain‑of‑thought reasoning step before generating robot actions. Using a hybrid decoder (autoregressive reasoning → bidirectional parallel action generation), the authors train on a newly curated embodied CoT dataset and then refine with outcome‑driven reinforcement learning. The model achieves 97 % average success on the LIBERO benchmark, runs at only 0.175× the latency of a pure autoregressive baseline, and transfers zero‑shot to the newer LIBERO Plus suite.

  • Project
    open-gigaai/giga-brain-0

    GigaBrain‑0.7 is an open‑source, large‑scale vision‑language‑action model for robots. It is pretrained on 37 k h of heterogeneous embodied data, supports eight robot embodiments, and ships with training scripts, benchmark suites (VLM‑Benchmark, RoboColiseum, RoboTwin 2.0, EBench), and ready‑to‑run inference servers that integrate with ROS.

  • Project
    opendilab/LightRFT

    LightRFT is an open‑source, distributed reinforcement‑learning framework for fine‑tuning large language and multimodal generative models. It unifies training back‑ends (DeepSpeed, FSDP) with rollout engines (SGLang, vLLM), supports a range of RL algorithms (GRPO, REINFORCE++, DAPO, etc.), and works across text, image, video, and audio modalities. The library provides ready‑to‑run examples (e.g., GSM8K with Qwen2.5), extensive docs, and Docker images, making it suitable for researchers and engineers building RL‑based alignment pipelines.

  • Project
    dsta022/Loop-Engineering-for-VLA

    Loop Engineering for VLAnything is an end‑to‑end Python toolkit that lets you record, merge, audit, enrich, and iteratively improve multimodal robot‑learning datasets (RGB, RGB‑D, and future visuo‑tactile) for vision‑language‑action models. It provides a conservative, side‑car‑first quality audit, plug‑in semantic evaluators, language‑annotation pipelines, and a closed‑loop feedback loop for policy improvement, all compatible with the LeRobot data format and publishable to Hugging Face.