We developed the framework of AutoOSS (Autonomous on-surface synthesis) to automate chemical reactions (breaking C-Br) in scanning tunneling microscopy. It comprises the remote connection, target dection module, interpetation module (image classifiers to identify reactants and products), decision-making module to optimize parameters as well as various analysis scritps.
- Clone the repository:
git clone https://github.com/SINGROUP/AutoOSS.git
- Navigate to the main directory：
cd AutoOSS - Install dependenceies：
pip install -r requirements.txt
conda install -c your-anaconda-username your-package-name
It consists of the interface to remote connection to STM/AFM software to monitor STM, target detection.
The reinforcement learning module aims to optimize manipulation parameters.
Neural network models based on ResNet18 can be applied to identify reactants and products, where bayesian optimization technique is used to optimize hyperparameters like learning rate.
The optimized neural network parameters of image classifiers were uploaded to evalute the protrusion in STM images.
It includes the script to show all dissociation cases with images and signal curves, to submit tasks, and all analyses in the manuscript.
Distributed under the MIT License. More details are shown in LICENSE.
If you use AutoOSS repository, please cite the following paper: