Nora-long Open-source Vision-Language-Action Model - Generate Robot Actions from Language Instructions and Images

Nora Long

Developed by declare-lab

A vision-language-action model trained on the Open X Embodied Dataset, generating robot motions from language instructions and camera images

Multimodal Fusion

Transformers

#Robot Motion Generation #Vision-Language-Action Model #5-step Action Span

Downloads 673

Release Time : 4/29/2025

Model Overview

Nora Long-range Version is an open-source vision-language-action model fine-tuned from Tongyi Qianwen 2.5 VL-3B, specifically designed for robotic manipulation tasks. It employs 5-step action span pre-training and demonstrates excellent performance in the LIBERO simulation environment

Model Features

Long-range Motion Prediction

Utilizes 5-step action span pre-training, suitable for tasks requiring long-term planning

Multimodal Input

Processes both language instructions and visual inputs simultaneously for more precise motion control

Open-source and Fine-tunable

Provides complete training code and model checkpoints, supporting user-customized fine-tuning

Model Capabilities

Vision-Language Understanding

Robot Motion Prediction

Multimodal Task Execution

Long-range Motion Planning

Use Cases

Robot Control

Robotic Arm Manipulation

Controls robotic arms to perform grasping, placing and other operations based on natural language instructions and visual inputs

Validated effective in WidowX robot tasks and LIBERO simulation environment

Automated Assembly

Completes complex assembly tasks through visual and language guidance

🚀 Nora-Long

Nora-Long is an open vision-language-action model that takes language instructions and camera images as input to generate robot actions, trained on the Open X-Embodiment dataset.

🚀 Quick Start

Nora-Long is an open vision-language-action model trained on robot manipulation episodes from the Open X-Embodiment dataset. It takes language instructions and camera images as input and generates robot actions. It is trained directly from Qwen 2.5 VL-3B. All Nora checkpoints, as well as our training codebase are released under an MIT License.

Unlike Nora, Nora-Long is pretrained with an action horizon of 5. We observe worse performance on the WidowX robot task with Nora-Long, but superior performance in the libero simulation. Please feel free to finetune this model!

✨ Features

Takes language instructions and camera images as input to generate robot actions.
Trained directly from Qwen 2.5 VL-3B.
Released under an MIT License.

📚 Documentation

Model Description

Model type: Vision-language-action (language, image => robot actions)
Language(s) (NLP): English
License: MIT
Finetuned from model: Qwen 2.5 VL-3B

Model Sources

Repository: https://github.com/declare-lab/nora
Paper: https://www.arxiv.org/abs/2504.19854
Demo: https://declare-lab.github.io/nora

💻 Usage Examples

Basic Usage

Nora takes a language instruction and a camera image of a robot workspace as input, and predicts (normalized) robot actions consisting of 7-DoF end-effector deltas of the form (x, y, z, roll, pitch, yaw, gripper). To execute on an actual robot platform, actions need to be un-normalized subject to statistics computed on a per-robot, per-dataset basis. Instructions on how to run Nora is available on https://github.com/declare-lab/nora.

Advanced Usage

You can finetune this model according to your specific needs. Since Nora-Long shows different performance in different scenarios (worse on WidowX robot task but superior in libero simulation), finetuning can potentially improve its performance in your target application.

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご