Multi-Tier 3D Trajectory Planning for Cellular-Connected UAVs in Complex Urban Environments

Xiling Luo, Tianyi Zhang, Wenxiang Xu, Chao Fang, Tongwei Lu, Jialiu ZhouSymmetry 2023 · DOI

TL;DR

This paper introduces a novel multi-tier trajectory planning (MTTP) method for cellular-connected UAVs in complex urban environments. By integrating air-ground communication assurance with advanced collision avoidance, MTTP significantly improves flight safety and mission success rates, outperforming existing communication-based approaches.

What Problem Does This Paper Address?

Unmanned Aerial Vehicles (UAVs) are increasingly used for various applications, but their widespread adoption in urban areas is hindered by communication and navigation challenges. Existing trajectory planning methods often struggle with the dynamic nature of urban environments, including sudden obstacles and unreliable communication links with ground base stations (GBSs). These limitations can lead to mission failures, safety risks, and inefficient flight paths. This paper specifically addresses the need for a robust trajectory planning system that can simultaneously ensure reliable communication and safe navigation for UAVs operating in complex 3D urban settings.

Methods

The proposed Multi-Tier Trajectory Planning (MTTP) method tackles this problem through a hierarchical approach:

  • Connectivity-Aware Global Planning Tier: This tier uses a Heuristic Search Algorithm (HSA) to find an optimal global path between GBSs. It models the urban environment as a directed graph where nodes are GBS coverage areas and edges represent voyages between them. The HSA considers both the average flight risk of GBS signal coverage and the distance to determine the best route, aiming for a globally optimal path that balances communication reliability and flight distance.
  • Collision-Free Local Planning Tier: Once a global path is established, this tier uses a Deep Reinforcement Learning (DRL) approach, specifically a Double Deep Q-Network (DDQN) with a dueling network structure (D3QN), to plan the detailed 3D trajectory within each GBS coverage area. This DRL agent learns to navigate complex environments by considering static obstacles, unexpected obstacles detected by sensors, and communication conditions. It employs a collaborative offline and online strategy to handle both known and dynamic threats.

The core innovation lies in the integration of these two tiers and the development of a flight risk model that accounts for both communication outage probability and environmental complexity. The problem is formulated as a minimal risk distance model.

Datasets

The paper utilizes simulated urban environments for experiments, including:

  • Virtual City Environment: A 500m x 500m x 25m area with 123 distributed buildings, 25 GBSs, and varying numbers of Temporary Flight Restrictions (TFRs) to simulate different levels of environmental complexity.
  • Realistic Urban Area: A real-world zone in Yuhang District, Hangzhou, with complex urban architecture, 14 GBSs, and 9 TFRs.

The paper does not introduce new benchmark datasets but rather uses these simulated and real-world environments to evaluate the proposed method against a communication-based approach.

Key Insights and Ideas

  • Hierarchical Planning is Crucial: Separating the problem into global (GBS-to-GBS) and local (within GBS coverage) planning tiers significantly simplifies the complex 3D trajectory optimization problem.
  • Flight Risk as a Unified Metric: Combining communication outage probability and environmental complexity into a single "flight risk" metric provides a more holistic approach to safe and reliable UAV navigation.
  • Synergy of HSA and DRL: The combination of a heuristic search for global pathfinding and DRL for local, dynamic path planning leverages the strengths of both approaches, leading to more efficient and robust solutions.
  • Asymmetrical Control is Key: The paper highlights the importance of asymmetrical control in cellular-connected UAVs, balancing communication needs with flight safety in intricate urban settings.

Key Results

  • Reduced Flight Distance: The MTTP method achieved a reduction in flight distance of at least 8.49% compared to the communication-based approach across various average flight risks.
  • Increased Mission Success Rate: The MTTP method demonstrated an impressive 10% increase in the mission success rate, particularly in environments with higher flight risks.
  • Superior Performance in Complex Environments: Simulations and real-world experiments showed that MTTP consistently outperforms the communication-based approach, especially in challenging scenarios with numerous obstacles and TFRs.
  • Faster Convergence: The D3QN algorithm used in the local planning tier showed quick convergence and stable reward values during training.

Limitations and Future Work

  • Constant UAV Speed: The current model assumes a constant UAV speed. Future work aims to dynamically adjust flight speeds based on communication interruption probability to further optimize performance.
  • Static Obstacle Modeling: The current model considers static obstacles and temporary flight restrictions. Future work could incorporate more dynamic and detailed environmental information, such as building Point-of-Interest (POI) indices and crowd distributions, to align the planning process with more realistic scenarios.
  • Energy Consumption: The paper acknowledges that energy consumption was not explicitly considered as an optimization variable. Future work will focus on incorporating this factor to improve overall mission efficiency and endurance.

Summary updated

BibTeX
@article{luo2023multi,
  author = {Xiling Luo and Tianyi Zhang and Wenxiang Xu and Chao Fang and Tongwei Lu and Jialiu Zhou},
  title = {{Multi-Tier 3D Trajectory Planning for Cellular-Connected UAVs in Complex Urban Environments}},
  journal = {Symmetry},
  year = {2023},
  doi = {10.3390/sym15091628},
}
Figure