A Hyperspectral Imaging Guided Robotic Grasping System,

This is the official code release of A Hyperspectral Imaging Guided Robotic Grasping System.

[paper] [project] [code] [Datasets] [CAD files]

## Environment

The complete deployment of the project includes the following components:

• Model Training and Inference
• Robotic Manipulation
• PRISM Control (Hyperspectral Camera, Motors)

Due to the windows required of the hyperspectral camera control interface, the project is developed on :

• Windows 10.

But the model training and inference can be run on any platform such as Ubuntu 20.04 (tested) that supports PyTorch.

Installation

1. Create conda environment and install pytorch

   This code is tested on Python 3.10.14 on Ubuntu 20.04 and Windows 10.

   conda create -n prism python=3.10
   conda activate prism
   # pytorch with cuda 11.8
   pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
   

2. Dependencies

   Install dependencies

   pip install joblib
   pip install tqdm
   pip install tensorboard
   pip install omegaconf
   pip install opencv-python
   pip install matplotlib 
   pip install scipy
   pip install scikit-learn
   pip install plantcv
   pip install spectral
   pip install numpy==1.26.4
   pip install h5py
   

Inference and Visualization

Only tested under the pycharms environment. please unclick the "Run with Python Console" and "view > Scientific Mode" option in the run configuration.

Run commands below to run the prism working animation:

python scripts/prism_animation.py

After finished, you shall see printed 3D affordance results w/ grasp and visualization at `run_realworld/gym_outputs/drawer_open/` like below:

Train and Test

You can modify the config parameter model_type in config/train.yaml to train the specific model.

python scripts/train.py

You can also run the test script to evaluate the trained model.

python scripts/test.py

Control Codes

All C++ device control codes are in the "c_device" folder. This includes control modules for Modbus devices, the Nachi robot, and the Specim linescan camera.

--c_device
    --libModbus
    --nachi  
    --specim 

Citation

If you find this work helpful, please consider citing:

@ARTICLE{11020724,
  author={Sun, Zheng and Dong, Zhipeng and Wang, Shixiong and Chu, Zhongyi and Chen, Fei},
  journal={IEEE Robotics and Automation Letters}, 
  title={A Hyperspectral Imaging Guided Robotic Grasping System}, 
  year={2025},
  volume={},
  number={},
  pages={1-8},
  keywords={Hyperspectral imaging;Robots;Grasping;Robot sensing systems;Sorting;Service robots;Cameras;Robot vision systems;Nonlinear distortion;Servomotors;Perception for Grasping and Manipulation;Software-Hardware Integration for Robot Systems;Grasping},
  doi={10.1109/LRA.2025.3575654}}