森林环境下无人机多模态感知轨迹规划方法

doi:10.16182/j.issn1004731x.joss.26-0536

摘要/Abstract

摘要：

面向森林防火林区巡检中树干密集遮挡、深度观测局部失效条件下的无人机低时延局部规划与安全避障需求，提出一种融合深度图、局部占据距离图、候选碰撞风险判别和执行前安全重排的学习式局部轨迹规划 方法。以深度图、局部占据距离图、运动状态和目标方向为输入，通过3×5角度运动基元并行生成候选五次多项式轨迹；训练阶段构造候选安全代价、排序监督与近碰撞风险标签，推理阶段结合在线激光雷达点云对候选轨迹进行安全复核与重排。局部占据距离图用于补充单一深度模态在树干遮挡、深度缺失和视场边界处缺少水平空间先验的问题，风险辅助头与在线复核用于提高候选选择的安全冗余。仿真结果表明：在6 m/s条件下，该方法在低密度和高密度森林场景中的成功率分别为81%和57%，碰撞率分别为2%和5%，平均规划时延分别为20.81和23.27 ms；相较深度单模态基线，成功率分别提高29和12个百分点。该方法能够在保持实时性的同时提升局部避障安全性与闭环执行稳定性。

关键词: 森林障碍环境, 无人机, 轨迹规划, 多模态感知, 安全重排, 碰撞风险

Abstract:

To address the requirements for low-latency local planning and safe obstacle avoidance in UAV forest-fire patrol under dense tree-trunk occlusion and locally invalid depth observations, this paper proposes a learning-based local trajectory planning method that fuses depth images, local occupancy distance maps, candidate-level collision-risk assessment, and pre-execution safety reranking. The method uses depth images, local occupancy distance maps, motion states, and target direction as inputs and generates 15 quintic-polynomial candidate trajectories in parallel using 3 × 5 angular motion primitives. During training, candidate safety costs, ranking supervision, and near-collision risk labels are constructed. During inference, online LiDAR point clouds are used to recheck and rerank the candidate trajectories before execution. The local occupancy distance map compensates for the lack of horizontal spatial priors in the depth-only modality under trunk occlusion, missing depth, and near field-of-view boundaries, while the auxiliary risk head and online rechecking improve the safety redundancy of candidate selection. Simulation results show that at 6 m/s, the proposed method achieves success rates of 81% and 57% in low- and high-density forest scenarios, with collision rates of 2% and 5% and average planning latencies of 20.81 and 23.27 ms, respectively. Compared with the depth-only baseline, the success rates increase by 29 and 12 percentage points, respectively. The method improves the safety of local obstacle avoidance and the stability of closed-loop execution while maintaining real-time performance.

Key words: forest obstacle environment, unmanned aerial vehicle, trajectory planning, multimodal perception, safety reranking, collision risk

中图分类号:

TP391.9

李德权,蒋家豪 . 森林环境下无人机多模态感知轨迹规划方法[J]. 系统仿真学报, 2026, 38(10): 2872-2886.

Li Dequan,Jiang Jiahao . Multimodal Perception-based Trajectory Planning Method for UAVs in Forest Environments[J]. Journal of System Simulation, 2026, 38(10): 2872-2886.

图/表 12

图1

表1

表2

图2

表3

图3

图4

图5

图6

图7

图8

表4

参考文献 36

[1]	张峰玲, 童红卫, 黄天来, 等. 基于混合遗传算法的无人机森林防火巡护路径研究[J]. 浙江林业科技, 2024, 44(5): 132-139.
	Zhang Fengling, Tong Hongwei, Huang Tianlai, et al. Research on UAV Forest Fire Prevention Patrol Path Based on Hybrid Genetic Algorithm[J]. Journal of Zhejiang Forestry Science and Technology, 2024, 44(5): 132-139.
[2]	张彪, 康峰, 许舒婷. 基于改进RRT与GA的多目标路径规划-以无人机林区巡检为例[J]. 北京林业大学学报, 2025, 47(4): 129-141.
	Zhang Biao, Kang Feng, Xu Shuting. Multi-objective Path Planning Based on Improved RRT and GA: Taking UAV Forest Inspection as an Example[J]. Journal of Beijing Forestry University, 2025, 47(4): 129-141.
[3]	Zhou Yuquan, Yan Li, Han Yaxi, et al. A Survey on the Key Technologies of UAV Motion Planning[J]. Drones, 2025, 9(3): 194.
[4]	胡克, 孙洪飞. 未知混动态环境下多无人机轨迹规划[J]. 智能系统学报, 2025, 20(2): 445-456.
	Hu Ke, Sun Hongfei. Trajectory Planning for Multi-drone in Unknown Mixed Dynamic Environments[J]. CAAI Transactions on Intelligent Systems, 2025, 20(2): 445-456.
[5]	陆则宇, 王瑶, 吴蔚楠, 等. 三维动态环境下的无人机集群双层航迹规划[J]. 系统工程与电子技术, 2026, 48(1): 132-143.
	Lu Zeyu, Wang Yao, Wu Weinan, et al. Two-layered Path Planning for Unmanned Aerial Vehicle Swarm in Three-dimensional Dynamic Environment[J]. Systems Engineering and Electronics, 2026, 48(1): 132-143.
[6]	宋超, 李波, 马云红, 等. 基于优化A*和MPC融合算法的三维无人机航迹规划[J]. 系统工程与电子技术, 2023, 45(12): 3995-4004.
	Song Chao, Li Bo, Ma Yunhong, et al. 3D UAV Trajectory Planning Based on Optimized A* and MPC Fusion Algorithm[J]. Systems Engineering and Electronics, 2023, 45(12): 3995-4004.
[7]	李高扬, 黎向锋, 赵康, 等. 全粒子推动野马优化算法的无人机三维路径规划[J]. 系统仿真学报, 2024, 36(3): 595-607.
	Li Gaoyang, Li Xiangfeng, Zhao Kang, et al. Three-dimensional Path Planning of UAV Based on All Particles Driving Wild Horse Optimizer Algorithm[J]. Journal of System Simulation, 2024, 36(3): 595-607.
[8]	桂雪琪, 李春涛. 基于视野和速度引导的无人机集群避障算法[J]. 系统仿真学报, 2024, 36(3): 545-554.
	Gui Xueqi, Li Chuntao. UAV Swarm Obstacle Avoidance Algorithm Based on Visual Field and Velocity Guidance[J]. Journal of System Simulation, 2024, 36(3): 545-554.
[9]	江好胜, 武芳芳, 黄泽贤, 等. 动态障碍物环境下多四旋翼轨迹规划与跟踪[J]. 系统仿真学报, 2025, 37(8): 2089-2102.
	Jiang Haosheng, Wu Fangfang, Huang Zexian, et al. Trajectory Planning and Tracking for Multi-quadcopter in Dynamic Obstacle Environments[J]. Journal of System Simulation, 2025, 37(8): 2089-2102.
[10]	Zhou Xin, Wang Zhepei, Ye Hongkai, et al. EGO-planner: An ESDF-free Gradient-based Local Planner for Quadrotors[J]. IEEE Robotics and Automation Letters, 2021, 6(2): 478-485.
[11]	Ye Hongkai, Zhou Xin, Wang Zhepei, et al. TGK-planner: An Efficient Topology Guided Kinodynamic Planner for Autonomous Quadrotors[J]. IEEE Robotics and Automation Letters, 2021, 6(2): 494-501.
[12]	Ren Yunfan, Cai Yixi, Zhu Fangcheng, et al. ROG-map: An Efficient Robocentric Occupancy Grid Map for Large-scene and High-resolution LiDAR-based Motion Planning[EB/OL]. (2023-02-28) [2026-05-12]. .
[13]	张森, 代强强. 改进型深度确定性策略梯度的无人机路径规划[J]. 系统仿真学报, 2025, 37(4): 875-881.
	Zhang Sen, Dai Qiangqiang. UAV Path Planning Based on Improved Deep Deterministic Policy Gradients[J]. Journal of System Simulation, 2025, 37(4): 875-881.
[14]	吕茂隆, 丁晨博, 韩浩然, 等. 基于深度强化学习的无人机自主感知-规划-控制策略[J]. 自动化学报, 2025, 51(6): 1305-1319.
	Maolong Lü, Ding Chenbo, Han Haoran, et al. Autonomous Perception-planning-control Strategy Based on Deep Reinforcement Learning for Unmanned Aerial Vehicles[J]. Acta Automatica Sinica, 2025, 51(6): 1305-1319.
[15]	Guglielmo Del Col, Karjalainen Väinö, Hakala Teemu, et al. Autonomous Drone Navigation in Forest Environments Using Deep Learning[C]//The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Göttingen: Copernicus Publications, 2025: 87-93.
[16]	Lei Bangsong, Hu Wei, Ren Zhaoxu, et al. DRL-based UAV Autonomous Navigation and Obstacle Avoidance with LiDAR and Depth Camera Fusion[J]. Aerospace, 2025, 12(9): 848.
[17]	Sun Tao, Gu Jiaojiao, Mou Junjie. UAV Autonomous Obstacle Avoidance via Causal Reinforcement Learning[J]. Displays, 2025, 87: 102966.
[18]	Wang Fei, Zhu Xiaoping, Zhou Zhou, et al. Deep-reinforcement-learning-based UAV Autonomous Navigation and Collision Avoidance in Unknown Environments[J]. Chinese Journal of Aeronautics, 2024, 37(3): 237-257.
[19]	Zhang Yuhang, Low K H, Chen Lü. Partially-observable Monocular Autonomous Navigation for UAV Through Deep Reinforcement Learning[C]//AIAA AVIATION 2023 Forum. Reston, VA, USA: AIAA, 2023: AIAA 2023-3813.
[20]	Yang Yongsheng, Hou Zhiwei, Chen Hongbo, et al. DRL-based Path Planner and Its Application in Real Quadrotor with LIDAR[J]. Journal of Intelligent and Robotic Systems, 2023, 107(3): 38.
[21]	Zhang Yuhang, Yan Chao, Xiao Jiaping, et al. NPE-DRL: Enhancing Perception Constrained Obstacle Avoidance with Nonexpert Policy Guided Reinforcement Learning[J]. IEEE Transactions on Artificial Intelligence, 2025, 6(1): 184-198.
[22]	Lu Junjie, Zhang Xuewei, Shen Hongming, et al. You Only Plan Once: A Learning-based One-stage Planner with Guidance Learning[J]. IEEE Robotics and Automation Letters, 2024, 9(7): 6083-6090.
[23]	Lu Junjie, Tian Bailing, Shen Hongming, et al. LPNet: A Reaction-based Local Planner for Autonomous Collision Avoidance Using Imitation Learning[J]. IEEE Robotics and Automation Letters, 2023, 8(11): 7058-7065.
[24]	Cao Hongyu, Lu Junjie, Zhang Xuewei, et al. YOPO-rally: A Sim-to-real Single-stage Planner for Off-road Terrain[EB/OL]. (2025-05-24) [2026-05-12]. .
[25]	Bhattacharya A, Rao N, Parikh D, et al. Vision Transformers for End-to-end Vision-based Quadrotor Obstacle Avoidance[EB/OL]. (2025-04-01) [2026-05-12]. .
[26]	Lee J, Rathod A, Goel K, et al. Rapid Quadrotor Navigation in Diverse Environments Using an Onboard Depth Camera[EB/OL]. (2024-11-07) [2026-05-12]. .
[27]	Song Yunlong, Shi Kexin, Penicka Robert, et al. Learning Perception-aware Agile Flight in Cluttered Environments[C]//2023 IEEE International Conference on Robotics and Automation (ICRA). Piscataway: IEEE, 2023: 1989-1995.
[28]	Penicka Robert, Song Yunlong, Kaufmann Elia, et al. Learning Minimum-time Flight in Cluttered Environments[J]. IEEE Robotics and Automation Letters, 2022, 7(3): 7209-7216.
[29]	Li Yude, Zhou Zhexuan, Li Huizhe, et al. Agile in the Face of Delay: Asynchronous End-to-end Learning for Real-world Aerial Navigation[EB/OL]. (2026-03-07) [2026-05-12]. .
[30]	Wei Junhao, Li Yanxiao, Yao Dexing, et al. SAGA: A Robust Self-attention and Goal-aware Anchor-based Planner for Safe UAV Autonomous Navigation[EB/OL]. (2026-05-04) [2026-05-14]. .
[31]	Zhou Boyu, Pan Jie, Gao Fei, et al. RAPTOR: Robust and Perception-aware Trajectory Replanning for Quadrotor Fast Flight[J]. IEEE Transactions on Robotics, 2021, 37(6): 1992-2009.
[32]	Xu Zhefan, Deng Di, Dong Yiping, et al. DPMPC-planner: A Real-time UAV Trajectory Planning Framework for Complex Static Environments with Dynamic Obstacles[C]//2022 International Conference on Robotics and Automation (ICRA). Piscataway: IEEE, 2022: 250-256.
[33]	Higgins J, Mohammad N, Bezzo N. A Model Predictive Path Integral Method for Fast, Proactive, and Uncertainty-aware UAV Planning in Cluttered Environments[C]//2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Piscataway: IEEE, 2023: 830-837.
[34]	Soria Enrica, Schiano Fabrizio, Floreano Dario. Distributed Predictive Drone Swarms in Cluttered Environments[J]. IEEE Robotics and Automation Letters, 2022, 7(1): 73-80.
[35]	Li Bo, Song Chao, Bai Shuangxia, et al. Multi-UAV Trajectory Planning During Cooperative Tracking Based on a Fusion Algorithm Integrating MPC and Standoff[J]. Drones, 2023, 7(3): 196.
[36]	Tordesillas J, How J P. Deep-PANTHER: Learning-based Perception-aware Trajectory Planner in Dynamic Environments[J]. IEEE Robotics and Automation Letters, 2023, 8(3): 1399-1406.

测试速度/(m/s)	轨迹时域/s
2	5.00
4	2.50
6	1.67
8	1.25
10	1.00

模块	输入/通道	实现	输出
深度图分支	1×96×160	单通道ResNet18，输出层1×1卷积到64维	64维深度特征
OccMap分支	1×100×100	插值到96×160后输入单通道ResNet18，输出64维	64维占据特征
状态变换	v 、 a 、 g	归一化后变换到各primitive frame	9维候选状态特征
融合与共享头	137	1×1 Conv:137→256→256，ReLU激活	候选级融合特征
输出头	256通道	轨迹头9维；代价头Softplus；风险头Sigmoid	p / v / a 、代价、风险概率
模型规模	—	完整模型参数量约22.52×10⁶	训练/推理共用

方法	障碍密度/(株/m²)	成功率/%(95%CI)	碰撞率/%(95%CI)	超时率/%	最小安全距离/m	平均规划时延/ms	平滑性
Baseline	0.033	52[42.32,61.54]	9[4.81,16.23]	39	0.326±0.086	11.35	109.53
E2E-DepthNet		45[35.61,54.76]	17[10.89,25.55]	38	0.299±0.107	8.49	127.92
MPC-Replan		64[54.24,72.73]	5[2.15,11.18]	31	0.388±0.081	31.83	106.98
Async-E2E-LiDAR		72[62.51,79.86]	7[3.43,13.75]	21	0.218±0.035	5.90	138.11
Ours		81[72.22,87.49]	2[0.55,7.00]	17	0.448±0.078	20.81	114.30
Baseline	0.050	45[35.61,54.76]	16[10.10,24.42]	39	0.304±0.110	13.77	131.23
E2E-DepthNet		37[28.18,46.78]	18[11.70,26.67]	45	0.293±0.114	11.03	125.90
MPC-Replan		49[39.42,58.65]	9[4.81,16.23]	42	0.327±0.081	34.32	107.31
Async-E2E-LiDAR		60[50.20,69.06]	12[7.00,19.81]	28	0.149±0.028	6.30	151.80
Ours		57[47.22,66.27]	5[2.15,11.18]	38	0.364±0.079	23.27	101.75

配置	模块组成	密度/(株/m²)	成功率/% (95%CI)	碰撞率/% (95%CI)	超时率/%	最小安全距离/m
A0	Depth	0.033	60[50.20,69.06]	14[8.53,22.14]	26	0.309±0.101
A1	Depth+Rank		52[42.32,61.54]	22[15.00,31.07]	26	0.293±0.120
A2	A1+OccMap		54[44.26,63.44]	14[8.53,22.14]	32	0.322±0.104
A3	A2+Collision Head		71[61.46,78.99]	8[4.11,15.00]	21	0.358±0.094
A4	A3+LiDAR Rerank		81[72.22,87.49]	2[0.55,7.00]	17	0.448±0.078
A0	Depth	0.050	43[33.73,52.78]	19[12.51,27.78]	38	0.297±0.126
A1	Depth+Rank		40[30.94,49.80]	25[17.55,34.30]	35	0.276±0.125
A2	A1+OccMap		51[41.35,60.58]	15[9.31,23.28]	34	0.304±0.103
A3	A2+Collision Head		53[43.29,62.49]	13[7.76,20.98]	34	0.312±0.095
A4	A3+LiDAR Rerank		57[47.22,66.27]	5[2.15,11.18]	38	0.364±0.079