An adaptive trajectory control method for unmanned aerial vehicles responding to traffic incidents at three-level interchange bridges

CHEN Deqi, ZHAO Meng, ZHANG Wenhui, YAN Xuedong, JIANG Xiancai

Journal of Tsinghua University(Science and Technology) ›› 2026, Vol. 66 ›› Issue (11) : 2270-2284.

PDF(2348 KB)
PDF(2348 KB)
Journal of Tsinghua University(Science and Technology) ›› 2026, Vol. 66 ›› Issue (11) : 2270-2284. DOI: 10.16511/j.cnki.qhdxxb.2026.26.066
OPERATIONAL SAFETY ASSURANCE FOR AUTONOMOUS ROAD TRAFFIC SYSTEMS

An adaptive trajectory control method for unmanned aerial vehicles responding to traffic incidents at three-level interchange bridges

  • {{article.zuoZhe_EN}}
Author information +
History +

Abstract

[Objective] Traffic incident response in autonomous road traffic systems requires a closed information loop involving roadside sensing, dispatch platforms, ground emergency forces, and low-altitude unmanned aerial vehicles (UAVs). In three-level interchange bridges, incidents may occur at different levels, while congestion, ramps, structural occlusion, and level misjudgment can delay ground access. After receiving target coordinates, a UAV must rapidly approach the assigned accident point and maintain stable trajectory tracking under wind disturbance and unreliable global positioning system (GPS) observations. This study focuses on the UAV execution layer and aims to improve the accuracy, smoothness, and robustness of adaptive trajectory control in complex interchange environments. [Methods] A PyBullet-based emergency flight environment was built for a three-level interchange bridge. The task was formulated as a continuous-control Markov decision process in which the UAV departed from a base station, followed waypoint sequences, and reached accident targets at different levels of the bridge. The state included estimated position and velocity, relative displacement to the current waypoint, attitude, angular velocity, acceleration, jerk, action variation rate, disturbance information, GPS status, and route index. The action space consisted of continuous thrust and attitude-channel inputs. Safety constraints and rewards jointly described boundary keeping, collision avoidance, target arrival, trajectory tracking, stability, energy use, and efficiency. Accordingly, a Kalman-(twin delayed deep deterministic policy gradient) TD3 method was proposed by embedding Kalman filtering into the online interaction process of the twin delayed deep deterministic policy gradient. The filter estimated the position, velocity, attitude, and angular velocity at each control step. The estimated position and velocity were used to reconstruct target-relative displacement, and stability variables were introduced into policy input and reward constraints. To adapt to uncertain conditions, process noise was adjusted according to wind intensity, and measurement noise was adjusted according to the magnitude of GPS anomalies. The performances of PPO, A2C, DDPG, TD3, and Kalman-TD3 were compared in disturbance-free, strong-wind, and GPS signal-loss scenarios. [Results] Kalman-TD3 achieved more stable training curves and smaller reward oscillations than the baseline algorithms, especially under strong wind and GPS anomalies. In the disturbance-free scenario, its task success rate reached 99.51%. Compared with TD3, Kalman-TD3 reduced the average tracking error (0.428 vs. 0.207), root mean square error (0.53 vs. 0.31), maximum error (0.97 vs. 0.64), and energy consumption (30.83 vs. 24.48). Under strong-wind conditions, it achieved a task success rate of 98.63%, an average tracking error of 0.246, a root mean square error of 0.36, and a maximum error of 0.72, while also reducing jerk and energy consumption relative to TD3. Under GPS signal loss, it maintained a task success rate of 96.43% and continued to outperform TD3 in average tracking error (0.499 vs. 0.252), root mean square error (0.61 vs. 0.37), and maximum error (1.12 vs. 0.74). Trajectory visualization further showed that Kalman-TD3 followed the planned waypoints more closely and changed levels more smoothly. [Conclusions] By integrating state estimation into deep reinforcement learning and adapting noise parameters to wind and GPS conditions, Kalman-TD3 reconstructs more reliable policy inputs from degraded observations. The method improves trajectory tracking accuracy, flight smoothness, energy efficiency, and robustness. In addition, it can support rapid incident access, on-site situation acquisition, and emergency dispatch assistance in autonomous road traffic systems.

Key words

unmanned aerial vehicle / traffic incident response / three-level interchange bridge / adaptive trajectory control / deep reinforcement learning / Kalman filtering

Cite this article

Download Citations
CHEN Deqi, ZHAO Meng, ZHANG Wenhui, YAN Xuedong, JIANG Xiancai. An adaptive trajectory control method for unmanned aerial vehicles responding to traffic incidents at three-level interchange bridges[J]. Journal of Tsinghua University(Science and Technology). 2026, 66(11): 2270-2284 https://doi.org/10.16511/j.cnki.qhdxxb.2026.26.066

References

[1] 王海波, 王子豪, 程华镇, 等. 基于改进Q-Learning算法的复杂风场中无人机三维路径规划[J/OL]. 华南理工大学学报(自然科学版). (2026-03-23) [2026-04-29]. https://link.cnki.net/urlid/44.1251.T.20260322.1356.004. WANG H B, WANG Z H, CHENG H Z, et al. Three-dimensional path planning for UAV in complex wind fields based on an improved Q-Learning algorithm[J/OL]. Journal of South China University of Technology (Natural Science Edition). (2026-03-23) [2026-04-29]. https://link.cnki.net/urlid/44.1251.T.20260322.1356.004. (in Chinese)
[2] 陈谋, 甄子涵, 姜斌. 低空无人机智能避障决策与安全控制技术综述[J]. 航空学报, 2026, 47(9): 533388. CHEN M, ZHEN Z H, JIANG B. Overview of intelligent obstacle avoidance decision-making and safety control technologies for low-altitude UAV[J]. Acta Aeronautica et Astronautica Sinica, 2026, 47(9): 533388. (in Chinese)
[3] 李立, 杨静宇, 张云, 等. 无人机道路交通安全巡查规划与事故现场勘察方法综述[J/OL]. 交通运输工程学报. (2026-04-01) [2026-04-29]. https://doi.org/10.19818/j.cnki.1671-1637.2026.285. LI L, YANG J Y, ZHANG Y, et al. A review of UAV path planning for road traffic safety inspection and accident scene investigation methods[J/OL]. Journal of Traffic and Transportation Engineering. (2026-04-01) [2026-04-29]. https://doi.org/10. 19818/j.cnki.1671-1637.2026.285. (in Chinese)
[4] 陈德启, 张自设, 张文会, 等. 面向高层建筑应急救援的无人机螺旋搜索轨迹控制方法[J]. 交通运输系统工程与信息, 2025, 25(6): 87-100. CHEN D Q, ZHANG Z S, ZHANG W H, et al. Trajectory control method for UAV spiral search oriented to high-rise building emergency rescue[J]. Journal of Transportation Systems Engineering and Information Technology, 2025, 25(6): 87-100. (in Chinese)
[5] 陈德启, 张淑慧, 张文会, 等. 面向林区公路高风险路段无人机自适应巡航轨迹规划方法[J]. 交通运输工程学报, 2026, 26(4): 121-133. CHEN D Q, ZHANG S H, ZHANG W H, et al. Method for UAV adaptive cruising trajectory planning for high-risk road sections of forest roads[J]. Journal of Traffic and Transportation Engineering, 2026, 26(4): 121-133. (in Chinese)
[6] 闫晓楠. 无人机在智慧交通中的应用[J]. 信息记录材料, 2026, 27(5): 81-83. YAN X N. Application of UAVs in intelligent transportation[J]. Information Recording Materials, 2026, 27(5): 81-83. (in Chinese)
[7] 滕菲, 王迎春, 姚永辉, 等. 基于深度强化学习的无人机动态避障规划[J/OL]. 北京航空航天大学学报. (2025-05-23) [2026-04-29]. https://doi.org/10.13700/j.bh.1001-5965.2025.0084. TENG F, WANG Y C, YAO Y H, et al. Dynamic obstacle avoidance planning for UAV based on deep reinforcement learning[J/OL]. Journal of Beijing University of Aeronautics and Astronautics. (2025-05-23) [2026-04-29]. https://doi.org/10.13700/j.bh.1001-5965.2025.0084. (in Chinese)
[8] 朱越, 王睿, 周洲. 基于深度强化学习的中小型无人机高原峡谷抗风飞行控制[J]. 西北工业大学学报, 2026, 44(1): 1-11. ZHU Y, WANG R, ZHOU Z. Wind disturbance-resilient flight control for small and medium-sized UAVs in plateau canyon environments using deep reinforcement learning[J]. Journal of Northwestern Polytechnical University, 2026, 44(1): 1-11. (in Chinese)
[9] TEOH J, LIM Z X, CHAN Y H, et al. Recent advancements, challenges, and future directions of integrating drones in smart city traffic management: A systematic review[J]. Journal of Electrical and Computer Engineering, 2026, 2026(1): 7869329.
[10] YUE P Y, XIN J, HUANG Y, et al. UAV autonomous navigation system based on air-ground collaboration in GPS-denied environments[J]. Drones, 2025, 9(6): 442.
[11] LIU S Y, ZHANG Y, LI G X, et al. A hierarchical adaptive navigation planner for UAVs in 3D complex environments based on deep reinforcement learning[J]. Applied Soft Computing, 2026, 190: 114614.
[12] CHENG C Q. Application of reinforcement learning to UAV decision making and path planning[J]. Frontiers in Computing and Intelligent Systems, 2025, 14(1): 30-33.
[13] AHMED A H, TOMÁN H. A hybrid framework for UAV obstacle avoidance integrating reactive sensing, LiDAR planning, and deep reinforcement learning in real time[J]. Results in Engineering, 2026, 30: 110150.
[14] ZHOU L, XU C, TIAN F N, et al. Control algorithms for quadrotor UAVs under wind disturbance[J]. Journal of Aerospace Engineering, 2026, 39(2): 04025131.
[15] CHANG Y X, CHENG Y Q, MANZOOR U, et al. A review of UAV autonomous navigation in GPS-denied environments[J]. Robotics and Autonomous Systems, 2023, 170: 104533.
[16] ERÜST A C, TAŞCIKARAOĞLU F Y, KÜÇÜKDEMIRAL İ B. Dynamic event-triggered robust model predictive control for quadrotor trajectory tracking[J]. International Journal of Robust and Nonlinear Control, 2026, 36(6): 3676-3688.
[17] DE SOUSA BEZERRA C D, VIEIRA F H T, DA SILVA SOARES A. Deep-Q-network hybridization with extended Kalman filter for accelerate learning in autonomous navigation with auxiliary security module[J]. Transactions on Emerging Telecommunications Technologies, 2024, 35(2): e4946.
[18] ZORAIN M, KHAN F S, HASANY N, et al. Deep reinforcement learning for UAV attitude control via adaptive gain optimization[J]. Applied Intelligence, 2025, 55(17): 1092.
[19] FAN M F, WANG G Z, CHEN H, et al. Deep reinforcement learning-based path planning for uncrewed systems: A survey[J/OL]. Unmanned Systems. (2026-04-07) [2026-09-01]. DOI: 10.1142/S2301385028300028.
[20] SONNY A, YEDURI S R, CENKERAMADDI L R. Q-learning- based unmanned aerial vehicle path planning with dynamic obstacle avoidance[J]. Applied Soft Computing, 2023, 147: 110773.
PDF(2348 KB)

Accesses

Citation

Detail

Sections
Recommended

/

〈 〉