This paper presents a novel hybrid CBS-PRM framework for efficient multi-robot path planning in congested environments. Addressing the NP-hard complexity of multi-agent navigation, our approach integrates conflict-based search with probabilistic roadmaps through three key innovations: adaptive rectangular sampling regions that concentrate nodes in critical pathways while avoiding obstacles; a directional expansion strategy resolving connectivity issues in narrow passages via potential field-guided optimization; and dynamic connection length adjustment responding to local environmental constraints. The methodology demonstrates significant improvements in computational efficiency and path quality while maintaining rigorous collision avoidance guarantees, providing an effective solution for warehouse automation and industrial robotics applications requiring coordinated multi-agent movement through constrained spaces.