DYNAMIC RISK WEIGHING IN HIERARCHICAL ENVIRONMENTS WITH MOVING OBSTACLES

Abstract

The paper addresses the critical problem of constructing safe and stable routes for autonomous agents in complex dynamic environments. The aim of the study is to improve the efficiency and intelligence of navigation by integrating a dynamic risk-weighting mechanism into a two-level hierarchical path planning model. Within the scope of this research, a mathematical model was developed to formalize the computation of traversal costs for hierarchical clusters based on the object movement intensity coefficient. This enables strategic safety assessment of routes at the high-level planning stage and facilitates the avoidance of potentially hazardous areas.

The research methodology is based on hierarchical space decomposition, the application of the any-angle pathfinding algorithm (Theta*) to ensure trajectory smoothness, and the implementation of the proposed mathematical model for real-time dynamic adjustment of graph weight coefficients.

It was established that the practical application of the proposed model in structured scenarios reduces the number of agent–obstacle collisions by a factor of 4.4, ensuring continuous motion and a high level of physical safety. Despite a moderate increase in single-path computation time (on average by 38–44%), the overall computational stability of the system is significantly improved due to a 68% reduction in the number of required replanning operations. The proposed approach minimizes redundant execution of resource-intensive collision handling, animation transitions, and behavioral reaction logic, which typically consume substantially more computational resources in game systems than strategic path planning itself.

The practical significance of the obtained results is confirmed by a software implementation in the Unity environment, demonstrating the algorithm’s capability to generate smooth, natural, and strategically justified trajectories in dynamic environments without relying on a static NavMesh. The results can be applied in the development of complex RTS games, crowd simulation systems, and autonomous robot control systems.