IR-Bot: an autonomous robotic system for real-time chemical mixture analysis via infrared spectroscopy and machine learning

Abstract

Rapid and accurate quantification of chemical mixtures is vital in autonomous chemical experimentation, which can serve as real-time feedback that guides decision-making and reduces resource consumption. Here, we present IR-Bot, an intelligent system that integrates robotic automation, infrared (IR) spectroscopy, quantum chemical simulations, and machine learning (ML) to enable autonomous, real-time mixture analysis. The autonomy is driven by a newly developed large-language-model-based IR Agent, working alongside a multi-agent robotic system to orchestrate theoretical IR calculations, experimental data acquisition, and ML-driven interpretation. Central to IR-Bot is the Alignment-Prediction approach: experimental spectra are aligned with simulated references, and a pre-trained prediction model then provides composition estimates. This fusion of pre-training, alignment, and prediction yields accurate, interpretable results for both binary and ternary mixtures, as demonstrated via a Suzuki coupling reaction. Moreover, IR-Bot’s explainable ML framework uncovers the specific vibrational modes guiding its predictions, offering deeper insights into fundamental chemical behaviour. By enabling rapid, reliable, and autonomous composition analysis, IR-Bot paves the way for a new generation of data-driven laboratory workflows capable of dynamic decision-making and real-time modification of experimental conditions.