Automated guided vehicles (AGVs) are widely used in warehouse and manufacturing facilities for point to point transportation of cargo. A typical objective in such facilities is space efficiency (store as many items as possible in a limited space) and throughput maximization. The two objectives are conflicting. Storing more items limits the available space for paths that can be used by the AGVs to transport goods within the facility, forcing the AGVs to take longer, more congested paths limiting the facility's throughput. Therefore, to maintain acceptable throughput, an efficient task allocation and vehicle coordination is required. In this paper we study the automation of such vehicles in a warehouse with a specific topology. In the topology considered, vehicle movement is extremely constrained making the overall system prone to deadlocks. Delays caused by deadlocks are significant and seriously affect operational performance of the warehouse. Our main objective is to derive an efficient task assignment and agent coordination policy such that the overall system throughput is maximized. We present a problem transformation derived by the constraints and we propose a specific policy which is scalable and requires minimal computational power therefore it can be applied in real time.