Abstract: As a pivotal domain in artificial intelligence, Multi-Agent Systems (MAS) have achieved remarkable progress in robotics, unmanned aerial vehicles, and other fields in recent years. However, with the expansion of application scenarios and the increasing complexity of operational environments, existing MAS face new challenges in collaborative ability, system resilience, and performance efficiency. This paper first traces the development of MAS and typical application scenarios, and then summarizes the existing problems into three core challenges: the conflict and trade-off between task efficiency and collision avoidance, topological deadlocks caused by resource conflicts, and nonlinear performance degradation due to system-scale expansion. Based on this classification framework, this paper further focuses on the research progress of the three types of problems, analyzes the theoretical limitations of existing methods, and finally provides a comprehensive analysis and outlook on the current problem - solving ideas and future development trends of MAS.