LabGuard: Grounding Natural-Language Laboratory Rules into Runtime Guards for Embodied Laboratory Agents

2026-06-29

AI Summary

This work addresses the lack of effective runtime enforcement of natural language safety rules for embodied scientific agents operating in dynamic experimental environments. The authors propose the first complete compilation pipeline that translates natural language laboratory safety rules into an executable intermediate representation, termed LabGuard-IR. This framework employs a grounding model, LabGuard-Grounder, to map rules into guardian constraints enforced at controller boundaries and integrates them into the ACT system for real-time monitoring. Evaluated on the newly introduced LabGuard-Bench annotated dataset, the method achieves a task-level F1 score of 79.4 on unseen rules, reduces unsafe event rates from 39.5% to 23.8%, and maintains an intervention rate below 0.5% without compromising task success, thereby significantly enhancing both safety and generalization of embodied agents.

Abstract

Scientific embodied agents are increasingly capable of carrying out laboratory procedures, but executing these procedures safely in dynamic laboratory environments remains challenging. Current safety approaches often overlook the intermediate step of transforming laboratory natural language, including safety rules, manuals, protocols, and standard operating procedures, into machine-checkable runtime constraints. We introduce LabGuard (Laboratory Guard), a language-to-execution safety suite that grounds natural-language laboratory rules into executable specifications and deploys them as runtime guards. LabGuard includes three core components: LabGuard-IR, which defines a typed executable representation; LabGuard-Bench, which provides 812 supervised annotations expanded from 203 seed laboratory rules; and LabGuard-Grounder, which maps natural-language laboratory rules into LabGuard-IR. The resulting IR instances are handled by the LabGuard Pipeline, which compiles them into runtime monitors and applies them at the controller boundary. Experiments show that LabGuard generalizes to unseen laboratory-rule sources, achieves 79.4 task-scope F1, and reduces unsafe events from 39.5% to 23.8% after monitor compilation. In LabUtopia, its runtime monitors integrate with ACT, keeping interventions below 0.5% while preserving task success.

Problem

• laboratory safety
• natural language grounding
• runtime guards
• embodied agents
• executable specifications

Innovation

• runtime guards
• natural-language grounding
• executable specifications
• laboratory safety
• embodied agents