Adaptive, delay-resilient labor allocation in construction using deep reinforcement learning for cost–delay trade-offs

Purpose
Effective labor management is a persistent challenge in construction projects due to uncertainties in worker availability, productivity fluctuations and potential schedule delays. This study aims to develop an intelligent decision-support framework that optimizes labor allocation under uncertainty using reinforcement learning (RL). The objective is to minimize overall project costs, including labor and delay penalties, while maintaining on-time completion through adaptive workforce planning.

Design/methodology/approach
A RL framework based on the Proximal Policy Optimization algorithm was implemented within a custom simulation environment reflecting a 25-month construction project in Thailand. Two labor types were modeled: daily workers (high productivity, uncertain attendance) and monthly workers (consistent availability, fixed salary). The RL agent learned optimal monthly hiring strategies to balance productivity and cost while responding to dynamic project conditions and uncertainties.

Findings
The proposed RL approach achieved up to 11.5% and 23.4% cost reductions compared to baseline strategies using only daily or only monthly labor, respectively. The RL agent successfully completed the project on time without incurring delay penalties, demonstrating its effectiveness in balancing cost and schedule risks. Results confirm the capability of RL to dynamically adapt labor allocation decisions in uncertain project environments.

Originality/value
This study is among the first to apply RL to construction labor management under uncertainty using real project data. It introduces a data-driven, adaptive decision-support tool for optimizing labor strategies in complex projects. The framework provides practical insights for construction managers and establishes a foundation for future research in AI-based, multi-objective workforce scheduling and cost-risk optimization.