AI-Assisted Lab Automation Control Framework

The ultimate goal of our work has been to automatically translate written laboratory protocols (including old, dog-eared printouts taped into lab notebooks) into fully automated scripts that can control all of the automation hardware needed to successfully run them. Starting from our library of code that runs a myriad of lab automation systems (developed over many years), and our proven lab scripting environment that coordinates the hardware, we have added an AI code generation layer that is capable of performing OCR on written protocols and translating them to finished automation scripts. The process includes the following: (1) Code generation LLM's are extended with our instrument APIs to generate protocol scripts. (2) The baseline generators are further enhanced with training as necessary for the specific protocols required. (3) A database of common lab instrument APIs is supplemented with custom driver development as needed for the particular protocols and applications. (4) The protocol scripts are generated, employing multiple AI models and an AI "Meta Layer" that is used to analyze validity and correctness of candidate scripts. (5) AI-generated animations representing the hardware are validated and simulations are run to further verify and validate the automation scripts. (6) Scripts are tested in a validation mode on the automation hardware and results analyzed to further refine the code, as needed. (7) Final working scripts are presented for final testing by lab personnel. Following this approach, we can reduce the time needed to automate complex protocols by more than an order of magnitude, while improving overall correctness and reliability of the final process.