This study presents µDropAI, an AI-assisted digital microfluidics (DMF) framework for multistate droplet control. It integrates a semantic segmentation model into a custom DMF system to recognize droplet states (unsplit, splitting, split, merging) and interactions for feedback control. The model achieves high accuracy (error rate < 0.63%) in recognizing droplets of varying colors and shapes, enabling autonomous droplet manipulation. The system improves the precision of volume control in droplet splitting, achieving a coefficient of variation (CV) of 2.74%. The open-source µDropAI framework is designed for compatibility with existing DMF devices and paves the way for integration with multimodal large language models for fully automated control.

Kunlun Guo, Zerui Song, Jiale Zhou, Bin Shen, Bingyong Yan, Zhen Guo, Huifeng Wang