A Multi-Agentic Digital Twin Framework for Human-in-the-loop Workflows

Abstract Traditional robot task planning struggles to recover from failures and changing requirements because it relies on predefined rules, labelled datasets, and limited state awareness. Digital twins provide real-time virtual representations of physical systems, while multi-agent systems support collaborative, stepwise reasoning. However, their integration for failure evaluation, requirement adaptation, and state-aware robot planning remains underexplored, particularly at the architectural level. This study proposes a multi-agentic digital twin framework for human-in-the-loop workflows, enabling bidirectional communication between machines, digital twins, and multi-agents. The framework is validated through robot task planning for an aerospace product-qualification metrology process. Results show that digital-twin integration adds 885-1,149 milliseconds to communication between robots and agents because continuous state updates are required. Multi-agent workflows need more human feedback than single-agent workflows to align outputs with predefined goals and mitigate hallucinations, highlighting the importance of human-in-the-loop orchestration. Datasets, logs, code, videos, results, and deployment instructions are available on GitHub 1 .