Safe and efficient multi-agent planning for human–integrated smart manufacturing

Modern smart manufacturing requires flexible and safe coordination between human operators and autonomous agents. Current multi-agent system planning often relies on reactive safety mechanisms, leading to inefficiencies and workflow interruptions. Existing approaches typically overlook runtime safety policies during plan generation, producing functionally valid but operationally inefficient plans. As a result, avoidable slowdowns and emergency stops occur when humans enter shared workspaces, reducing both throughput and predictability. This research introduces a safety-aware planning approach that builds on an enhanced agent ontology by integrating agent negotiation with automated planning. A dynamic map of enhanced edges is constructed, where each edge represents a sequence of skills and its aggregated runtime execution cost. During runtime, agents negotiate to exclude unsafe edges and update costs based on runtime conditions, enabling the solver to generate plans that are both safe and efficient. Evaluation in a Plug & Produce case study with two system configurations demonstrated consistent advantages over a baseline reactive safety policy. The proposed planner, increased throughput by 50–80%, reduced execution costs by 20–55%, and lowered variability by up to a factor of 17. By integrating safety awareness into the planning stage, the approach improves efficiency, robustness, and compliance with human safety standards while supporting adaptable reconfigurable manufacturing. This work advances the vision of a human-centric future manufacturing by demonstrating that smart manufacturing systems can reason about operator presence to avoid hazards without compromising productivity.