Dual-Agent Framework for Cross-Model Verified Translation of Natural-Language Protocols into Robotic Laboratory Platform

Biological experiment protocols are typically written in unstructured natural language, while robotic laboratory platforms require precise, structured control commands. This semantic gap creates a significant bottleneck in laboratory automation. The authors sought to bridge this gap by creating an autonomous, agent-based framework capable of interpreting natural-language instructions and converting them into reliable, executable sequences for a robotic microplate-based platform.

The researchers proposed a dual-agent pipeline consisting of three primary stages:

1. Parser Agent: An LLM-based module that extracts experimental steps from natural-language text and organizes them into a structured format, explicitly separating manual user tasks from instrument-executable tasks.
2. Rule-Based Mapping Engine: A deterministic layer that converts the structured protocol into specific device-level control commands. This layer incorporates physical constraints, such as well-mapping, tip-usage policies, and coordinate transformations, ensuring that the generated commands are physically feasible.
3. Validation Agent: A heterogeneous LLM-based validator that checks the generated command sequence for completeness, parameter accuracy, and logical ordering. If errors are detected, the system triggers a self-correction loop to refine the output.

The framework was evaluated using a sweep of 7 different parsers and 3 validators on ELISA protocols and was further validated through the end-to-end execution of a Bradford assay on the KIMM BioForge-1 robotic platform.

This framework addresses the "hallucination" problem common in end-to-end LLM-based automation by separating high-level semantic interpretation from low-level deterministic execution. By using a rule-based engine to handle physical constraints, the system ensures that the robotic platform operates within its safety and operational limits. This approach shifts the researcher's role from writing manual scripts to designing experiments, potentially increasing the throughput and reproducibility of high-throughput biological assays.