A Bidirectional Feedback Actor–Critic Approach for Collaborative Target Search in Heterogeneous Intelligent Agent Systems

In target search tasks such as post-disaster rescue, collaboration between unmanned aerial vehicles and unmanned ground vehicles can significantly improve search efficiency. This study proposes an air–ground heterogeneous unmanned system cooperative search method termed bidirectional feedback actor–critic (BiF-AC). The key innovation lies in integrating aerial wide-area coverage with precise ground positioning to design a point-area combined search mechanism. A dynamic Voronoi-based method is further adopted to enhance collaboration within heterogeneous systems. Considering the characteristics of the target search problem, an actor–critic framework is introduced for agent training, bidirectional information interaction is employed, and a multilevel feedback reward function is constructed to ensure efficient search performance in diverse environments. Experimental results show that BiF-AC improves the target search success rates by an average of 40.3% and reduces search steps by 21.3% compared with other baseline algorithms. This research offers theoretical support and implementation paths for intelligent unmanned systems in emergency search, environmental monitoring, and related fields.