ChemOS is a software package developed by the Aspuru-Guzik Group at the University of Toronto designed to supply the structured layers indispensable for the deployment and operation of autonomous or "self-driving" laboratories (SDLs).
Its primary goal is to democratize and accelerate scientific discovery in chemistry and material science by bridging the gap between automated robotic platforms, artificial intelligence, and high-performance computing. The software enables a closed-loop approach to experimentation, where experiments are iteratively designed on-the-fly by a machine-driven decision-maker based on previously conducted experiments.
ChemOS follows a modular architecture composed of a central workflow manager and six independent modules:
- AI Algorithms for Experiment Planning (Learning Module): Incorporates state-of-the-art machine learning methods, such as Bayesian optimization algorithms like Phoenics and SMAC, to suggest optimal experimental conditions.
- Automation and Robotics: Manages the execution of experiments on automated robotic platforms.
- Characterization Equipment: Integrates with equipment to assess the performance of conducted experiments and provide feedback.
- Databases Handling and Management: Coordinates data collection and long-term data storage, supporting systems like SQLite.
- Intuitive Interfaces for Researchers: Facilitates interaction with researchers, including an NLP module in a chatbot framework for remote command and instruction.
- Online Results Analysis: Features an analysis module for pre-processing, summarizing, and visualizing experimental results, including time traces and search space visualization.
This modularity decouples interdependent tasks, allowing for easy extension and simplified deployment to new applications by only modifying a configuration file. It also supports remote control of laboratories and access to distributed computing resources.