Lean post-typhoon recovery: multi-agent look-ahead and constraints for the Last Planner System

Post-typhoon recovery involves high-variability work in which plan reliability is frequently undermined by late discovery of constraints, fragmented information, and rework. Rapid post-typhoon building recovery requires integrating heterogeneous evidence and converting it into actionable and auditable plans. This paper proposes a Lean Construction informed, knowledge-driven framework that combines a multi-modal knowledge graph (MMKG) with multi-agent orchestration for end-to-end building damage assessment and recovery planning, aligned with the Last Planner System (LPS) through look-ahead planning, make-ready checks, and explicit constraint management. The MMKG captures emergency management criteria, typhoon evolution and exposure status, building-level damage evidence, and reusable post-disaster recovery templates. The approach uses the MMKG as a shared and traceable evidence base, enabling multi-agent orchestration to generate a constraint log, assign make-ready status, and support weekly work plan release under explicit rule-based conditions. Typhoon Yagi is used as an illustrative proof-of-concept case to demonstrate how heterogeneous post-typhoon evidence can be translated into LPS-oriented planning artifacts. The results indicate that the framework is feasible for generating traceable planning outputs under multi-source uncertainty; however, quantitative field-scale validation and execution stage performance assessment remain future work.