README.md

metadata

language:
  - en
tags:
  - embeddings
  - multimodal
  - olfaction-vision-language
  - olfaction
  - olfactory
  - diffusion
  - scentience
  - neural-network
  - graph-neural-network
  - gnn
  - molecule
  - vision-language
  - vision
  - language
  - robotics
  - multimodal
  - smell
license: mit
datasets:
  - kordelfrance/olfaction-vision-language-dataset
  - detection-datasets/coco
base_model: DEGNN-Unconstrained

Diffusion Graph Neural Networks for Robust Olfactory Navigation in Robotics

Olfaction • Vision • Language

An open-sourced diffusion-based equivariant graph neural network (DEGNN) for olfaction-vision-language tasks.

---

Model Description

Navigation by scent is a capability in robotic systems that is rising in demand. However, current methods often suffer from ambiguities, particularly when robots misattribute odours to incorrect objects due to limitations in olfactory datasets and sensor resolutions. To address challenges in olfactory navigation, we introduce a novel machine learning method using diffusion-based molecular gen- eration that can be used by itself or with automated olfactory dataset construction pipelines. Our models, diffusion-based equivariant graph neural networks (DEGNN for short), leverage the state of the art in molecular generation and aroma mapping. This generative process of our diffusion model expands the chemical space beyond the limitations of both current olfactory datasets and training methods, enabling the identification of potential odourant molecules not previously documented. The generated molecules can then be more accurately validated using advanced olfactory sensors, enabling them to detect more compounds and inform better hardware design. By integrating visual analysis, language processing, and molecular generation, our framework enhances the ability of olfaction-vision models on robots to accurately associate odours with their correct sources, thereby improving navigation and decision-making through better sensor selection for a target compound in critical applications such as explosives detection, narcotics screening, and search and rescue. Our methodology represents a foundational advancement in the field of artificial olfaction, offering a scalable solution to challenges posed by limited olfactory data and sensor ambiguities.

We offer two models with this repository:

• (1) DEGNN-constrained: A diffusion model with its associated olfactory conditioner that is constrained to only generate molecules based on the atoms C, N, O, F, P, S, and Cl.
• (2) DEGNN-unconstrained: A diffusion model with its associated olfactory conditioner that is unconstrained and may generate molecules from any atom.

---

Model Details

• Model Name: DEGNN Constrained
• Developed by: Kordel K. France
• Date: September 2025
• Architecture:
  • Olfaction conditioner: Feedforward Neural Network
  • Diffusion model: Equivariant Graph Neural Network conditioned on atoms C, N, O, F, P, S, Cl
• License: MIT
• Contact: kordel@scentience.ai, kordel.france@utdallas.edu

---

• Model Name: DEGNN Unonstrained
• Developed by: Kordel K. France
• Date: September 2025
• Architecture:
  • Olfaction conditioner: Feedforward Neural Network
  • Diffusion model: Equivariant Graph Neural Network conditioned on all available atoms in training data
• License: MIT
• Contact: kordel@scentience.ai, kordel.france@utdallas.edu

---

Intended Use

• Primary purpose: Research in multimodal machine learning involving olfaction, vision, and language.
• Example applications:
  • Robotics and UAV navigation guided by chemical cues
  • Chemical dataset exploration and visualization
• Intended users: Researchers, developers, and educators working in ML, robotics, chemistry, and HCI.
• Out of scope: Not intended for safety-critical tasks (e.g., gas leak detection, medical diagnosis, or regulatory use).

---

Training Data

• Olfaction data: Language-aligned olfactory data curated from GoodScents and LeffingWell datasets.
• Vision data: COCO dataset.
• Language data: Smell descriptors and text annotations curated from literature.

For more information on how the training data was accumulated, please see the HuggingFace dataset URL here

---

Directory Structure

DiffusionGraphOlfactionModels/
├── data/                     # Example dataset
├── src/                      # Model training and inferenct tools
├── notebooks/                # Colab-ready notebooks
├── models/                   # Pre-trained models for immediate use
├── requirements.txt          # Python dependencies
├── LICENSE                   # Licensing terms of this repository
└── README.md                 # Overview of repository contributions and usage

---

Getting Started

The easiest way to get started is to open the Colab notebook and begin there. To explore the model and train locally, follow the steps below:

1. Clone the Repository

git clone https://github.com/KordelFranceTech/Diffusion-Graph-Olfaction-Models.git
cd DiffusionGraphOlfactionModels

2. Create a Virtual Environment

python -m venv env
source env/bin/activate  # On Windows: .\env\Scripts\activate

3. Install Dependencies

pip install -r requirements.txt

4. Run Inference or Train Models

Run inference:

python scripts/main.py

Train Models:

jupyter notebook notebooks/Olfaction_Diffusion-Train.ipynb

---

Citation

If you use these models in your research, please cite as follows:

@misc{france2025diffusiongraphneuralnetworks,
      title={Diffusion Graph Neural Networks for Robustness in Olfaction Sensors and Datasets}, 
      author={Kordel K. France and Ovidiu Daescu},
      year={2025},
      eprint={2506.00455},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2506.00455}, 
}

---

License

This dataset is released under the MIT License.