Self-driving laboratories increasingly rely on low-cost liquid handlers such as the Opentrons OT-2, which typically operate in an open-loop manner without pressure-based aspiration monitoring like those found in Hamilton or Tecan systems. This leads to two undetected failure modes: protocols that are syntactically valid but violate assay-specific invariants (e.g., tip reuse between a PCR template and a no-template control), and physical execution failures (partial dispense, air bubbles, missing tips) at runtime.
We present AEGIS, a two-layer guardian for both issues.
Layer 1 combines a curated machine-readable assay rule database with an LLM that reasons over OT-2 Python code, achieving an adjusted F1 of 0.97 on a benchmark of 24 protocols across five assay families, outperforming rules-only and LLM-only ablations. Notably, a free open-weight model matches the performance of the best proprietary model, eliminating the need for a paid API.
Layer 2 fits a PCA world model to YOLO-cropped four-frame pipette trajectories; under a leakage-free leave-one-plate-out evaluation, it achieves an average precision of 0.89 and an operating-point F1 of 0.71 (AUROC 0.80), matching the numbers from live demonstrations. We characterize the small-pipette (p20) resolution limit with an F1 of 0.47.
A live demonstration on a physical OT-2 (five replicates per condition) successfully detects planted no-tip failures and partial dispenses on colored dyes, with an always-VLM self-vote gate boosting partial-dispense recall to 5/5. Transparent water presents a principled limit for any front-view-only monitor, which AEGIS identifies as low-confidence VLM reasoning rather than a false verdict. Cascade triage keeps VLM costs at about $1.63 per plate compared to $10.33 for an always-VLM baseline. AEGIS is open-source and, to our knowledge, the first system to unify pre-flight assay-aware validation with runtime visual monitoring for an open-source liquid handler.
Blogger's Review: The innovation of AEGIS lies in its effective integration of assay rule validation and real-time monitoring, enhancing the reliability of liquid handling robots. This open-source solution not only lowers the barrier to entry but also lays a solid foundation for the future of self-driving laboratories, showcasing immense potential for application and scalability.