AEGIS Validates and Monitors Open-Source Liquid Handling Robots

AEGIS is an innovative two-layer system designed to enhance the reliability and safety of low-cost, open-source liquid handling robots, such as the Opentrons OT-2. These robots typically lack the advanced aspiration monitoring found in more expensive systems, leaving them vulnerable to undetected protocol and execution failures. AEGIS addresses this by providing both pre-flight validation and real-time monitoring. The first layer of AEGIS performs assay-aware protocol validation. It combines a curated database of machine-readable assay rules with a large language model (LLM) that analyzes OT-2 Python code. This layer achieves high accuracy (adjusted F1 of 0.97) in detecting protocol violations across various assay families, outperforming rule-only or LLM-only approaches. Notably, it can use a free open-weight LLM, eliminating the need for proprietary API costs. The second layer focuses on runtime monitoring of physical execution. It employs a PCA world model applied to YOLO-cropped pipette trajectories from four-frame video, achieving an average precision of 0.89 in detecting failures like partial dispenses or missing tips. A live demonstration confirmed its ability to deterministically catch no-tip failures and partial dispenses with colored dyes. The system also incorporates a VLM self-vote gate to improve recall for partial dispenses and manages VLM costs efficiently through cascade triage. AEGIS is open-source and represents the first unified solution for both validation and monitoring in this domain.