Simulating travel paths of construction site workers via deep reinforcement learning considering their spatial cognition and wayfinding behavior

Many optimization methods for construction site layout planning (CSLP) generate the shortest path of workers to calculate traveling costs and site safety performance. However, this approach often degrades the solution's reliability because workers in real-life situations do not necessarily take the shortest path to their chosen destination. Thus, this paper proposes a novel approach for generating realistic paths that mimic their way-finding decision-making process. This approach uses deep reinforcement learning, for which the framework to facilitate its use includes the following elements: (1) the required properties and functions for site objects; and (2) the state, action space, and reward functions intended. The similarity between the paths simulated and the real workers' trajectories has been validated better by 17.8% than the traditional A* algorithm. The proposed approach is expected to be used as an appropriate input, and thereby help improve the reliability of the solutions based on the CSLP optimization methods.

Automated summary

This study proposes a novel approach for generating realistic paths by simulating workers' way-finding decision-making process using deep reinforcement learning. The results show that the proposed approach simulates workers' trajectories better than the traditional A* algorithm.