A Real-Time System for Scheduling and Managing UAV Delivery in Urban Areas

1. Problem Statement

The rapid growth of urban logistics has made Unmanned Aerial Vehicle (UAV) delivery a promising solution for improving efficiency and reducing traffic congestion. However, current research often faces a critical gap between high-level scheduling algorithms and low-level execution systems.

• The Disconnect: Many existing scheduling models rely on unrealistic simplifications, are designed for offline use (causing latency), and fail to translate abstract scheduling requirements into specific, real-time execution commands.
• The Challenge: There is a lack of a middleware system that can bridge this gap, ensuring that scheduling algorithms are aware of the exact state of the underlying infrastructure (UAVs, ground vehicles, and staff) and can dynamically adjust to real-world constraints like battery swapping, cargo loading, and collision avoidance.

2. Methodology

The authors propose a Real-Time Scheduling and Management System based on the "Airport-Unloading Station" model. In this model, a centralized UAV airport serves as a distribution center where goods are loaded, and UAVs fly to specific unloading stations for customer pickup.

A. System Architecture (Middleware)

The system acts as middleware connecting the scheduling layer to the execution layer, composed of three core nodes operating as independent real-time processes:

1. Master Management Node: The central hub that connects the scheduler with execution. It aggregates status updates from UAVs and Automated Guided Vehicles (AGVs) and distributes scheduling decisions.
2. UAV Management Node: Manages individual UAVs using Finite State Machines (FSMs) and multithreading.
3. AGV Management Node: Manages AGVs (which transport UAVs between ground work areas and aviation zones) using similar FSMs.

Key Technical Features:

• Distributed Architecture: Each vehicle runs on an independent thread, ensuring fault isolation and scalability.
• Finite State Machines (FSMs):
  • UAV States: Ready $\to$ On Car $\to$ Waiting Go $\to$ Flying Go $\to$ Waiting Back $\to$ Flying Back.
  • AGV States: Waiting Pickup $\to$ Waiting Go GW $\to$ Waiting Working $\to$ Waiting Go AW.
  • These FSMs handle state transitions based on conditions (e.g., "Landed," "On Car," "Have Cargo") and commands (e.g., "Delivery," "Load Cargo").

B. Collaborative Scheduling Schemes

The system implements three distinct scheduling schemes to coordinate AGVs, UAVs, and ground staff, optimizing the flow between the Aviation Work (AW) zone (takeoff/landing) and the Ground Work (GW) zone (loading/battery swap):

1. One-Cycle: A single loop with one takeoff point and one landing point for all AGVs.
2. Two-Cycle: Two loops sharing one takeoff point but with two separate landing points.
3. Three-Cycle: Three independent loops, each with dedicated takeoff and landing points.

C. Traffic Schedulers

• Ground Traffic Scheduler: Manages AGV movement to ensure a continuous assembly-line flow, minimizing UAV wait times at takeoff/landing points.
• Air Traffic Scheduler: Controls takeoff and landing permissions. It enforces safety constraints by:
  • Ensuring a minimum time gap ($t_{gap}$) between UAV arrivals at unloading stations to prevent conflicts.
  • Verifying that an AGV is available at the landing point before the UAV returns ($T_{land}(AGV) < T_{land}(UAV)$).

3. Key Contributions

1. Middleware Design: Developed a real-time management system that successfully bridges the gap between theoretical scheduling algorithms and physical execution, translating high-level plans into low-level commands.
2. FSM-Based Control: Implemented robust, state-machine-driven control logic for both UAVs and AGVs, ensuring precise, autonomous, and responsive task execution.
3. Collaborative Schemes: Proposed and evaluated three distinct collaborative scheduling strategies (One, Two, and Three-Cycle) to optimize the coordination between air and ground assets.
4. Validation: Provided a comprehensive experimental framework using Docker containers and ROS (Robot Operating System) to simulate realistic urban delivery scenarios.

4. Experimental Results

The system was tested in a simulation environment (Intel i7, 32GB RAM) using Docker modules for AGV control, UAV flight control (PX4-like), scenario management, and scheduling.

• Performance Metrics: Evaluated based on the number of delivered goods and a scoring function (100 points for on-time delivery, penalized for delays or overruns).
• Impact of UAV Count: Increasing the number of UAVs (from 4 to 16) increased total deliveries and scores, but with diminishing returns.
• Scheme Comparison:
  • Two-Cycle Scheme: Achieved the highest delivery scores and the most efficient balance. It maximized ground staff utilization (leading to more cargo loaded) while maintaining moderate AGV usage.
  • Three-Cycle Scheme: Resulted in the lowest scores. While it reduced waiting times, AGVs spent excessive time traveling between zones, leading to high AGV busy ratios but inefficient overall throughput.
  • One-Cycle Scheme: Suffered from long waiting times, resulting in low AGV utilization.
• Resource Utilization: The study highlighted that the Two-Cycle scheme offers the optimal trade-off, balancing ground staff workload and AGV travel time to maximize efficiency.

5. Significance

• Practical Applicability: This work moves beyond theoretical optimization by providing a complete, executable solution for urban UAV delivery, addressing the often-overlooked "last mile" of system integration.
• Scalability and Safety: The distributed, FSM-based architecture ensures the system is scalable (easy to add new vehicles) and safe (strict state checks prevent collisions and execution errors).
• Operational Insight: The comparative analysis of the three scheduling schemes provides actionable guidelines for logistics companies, demonstrating that simply adding more vehicles or loops does not guarantee efficiency; the design of the workflow (loops and zones) is critical.
• Commercial Viability: By validating the system through extensive simulation with realistic parameters (battery swap times, loading times, speeds), the paper offers a viable blueprint for commercial drone delivery networks.

Code Availability: The authors have open-sourced the system at https://github.com/chengji253/UAVDeliverySystem.