Verification-gated agentic mission-state governance for intelligent industrial multi-robot systems

Abstract

Agentic artificial intelligence can decompose industrial tasks, propose robot actions, and adapt execution plans. However, generated proposals may conflict with task dependencies, resource ownership, protective holds, or repair boundaries during long-horizon multi-robot execution. To govern these transitions, we present a proposal-source-agnostic mission-state framework that maintains an evolving task forest and a governed blackboard as the canonical mission state. A derived execution coupling topology exposes cross-branch dependencies, while a centralized serial coordinator verifies typed proposals before commitment. Across 30-seed Medium and Large benchmarks, mean completion is 0.921 and 0.898, respectively, higher than shared-interface implementations of the core DART-DAG and LiP-LP decision logic. Global repair achieves comparable completion but permits broader state modification. Under clean authoritative records, verification preserves completion relative to direct commitment and reduces accepted commits identified as invalid by represented-state checks from 0.332–0.395 per 100 commits to zero on Medium. The ten-seed Large subset shows the same numerical pattern, with p Holm = 0.054 , just above the 0.05 threshold. Together, these results support consistent, auditable, and localized mission-state updates.

The authors' abstract, as published at the source. Engineering Applications of Artificial Intelligence, 2026 · DOI ↗