Abstract: The road planning at the construction site is subject to multiple constraints of terrain undulation, foundation pit forbidden area and building obstacles. The traditional empirical layout is difficult to take into account the traffic efficiency and workers' path habits, and a quantifiable group behavior optimization method is urgently needed. In this paper, a virtual environment including three-dimensional terrain, forbidden foundation pit and building obstacles is constructed, and a multi-weight cost function is established with distance, slope and turning cost to drive worker agents with different travel preferences. The A* algorithm is used to plan the path of the agent between the specified starting and ending points, and the pass frequency heat map is generated by superposition. According to the principle of 'inverse function of heat map-lowest cumulative cost of frequency', an optimal recommended path is selected from the group path. The simulation results show that the system can complete the path simulation of 100 agents within 5 s and output the optimal recommended road. The road coincides with the high-frequency traffic area of the group by 84.3%, which is significantly better than the single shortest path scheme. The proposed multi-agent path simulation optimization method improves the path efficiency while taking into account the workers' habits, and provides quantitative decision support for the economic and efficient planning of the construction site road.