To enhance unmanned ground vehicle (UGV) intelligence in smart cities, disaster rescue, and infrastructure inspection, this paper investigates the collaborative optimization of multimodal fusion end-to-end architectures. Through dynamic alignment of heterogeneous features, a multi-head distillation attention mechanism, and parallel decision-planning, a high-precision, low-latency closed-loop autonomous navigation framework is constructed. A Multi-Head Distillation Attention-based Trajectory Prediction (MDA-TP) method is proposed, combining a BEVFormer-based multimodal fusion perception model with a two-stage progressive knowledge distillation framework. On NuScenes, the method achieves an ADE of 0.88 m, an FDE of 1.32 m (4.34% and 10.81% reductions), and a collision rate of 15.2%, with 42.6 M parameters and 46 ms latency. Ablation shows removing attention distillation increases FDE by 13.6%. For system validation, a multi-sensor UGV platform is built. Through NuScenes online testing and real-world closed-loop validation, the Euclidean deviation remains within 0.5 m. Compared with traditional distillation, speed prediction MSE is reduced by 51.5%, wheel angle RMSE by 58.4%, and route completion improves from 60.99% to 97.26%. The results provide practical support for autonomous driving in smart cities, disaster rescue, and infrastructure inspection.
Infrared (IR) video target detection is important for long-range target perception in anti-UAV systems under complex lighting and background conditions. However, long-range tiny unmanned aerial vehicles (UAVs) usually occupy only a few pixels, exhibit weak thermal contrast, and are easily disturbed by cluttered backgrounds such as clouds, buildings, vegetation, feature edges, and thermal noise. Although YOLO-style detectors provide high real-time performance, their feature flow is prone to weak target response attenuation during downsampling, contextual ambiguity in deep feature representation, and background clutter propagation during cross-scale fusion. To address these degradation problems, this paper proposes FDR-YOLO, a feature-degradation-aware feature-flow reconstruction network based on YOLOv26. Specifically, LAE-based response-preserving downsampling (LAE-RPD) preserves weak but discriminative target responses during spatial compression; lightweight U-shaped dilated context aggregation (UCDC-Lite) enhances deep contextual discrimination between tiny UAV targets and cluttered backgrounds; and high-frequency prior-guided semantic injection fusion (HPG-SIF) uses shallow high-frequency priors to constrain the injection of deep semantic features. Experiments on multiple datasets show that FDR-YOLO improves detection accuracy while retaining lightweight and low-latency characteristics. On the Anti-UAV dataset, FDR-YOLO improves mAP50 and mAP50–95 by 3.0 and 3.3 percentage points, respectively, over YOLOv26s. Additional experiments on InfraredUAV and the RGB-based UAVSwarm dataset demonstrate the applicability of the proposed design to another infrared benchmark and to visible-light UAV detection under dataset-specific training.
Complex urban scenarios impose high demands on the environmental perception capabilities of unmanned systems, which serve as a prerequisite for executing autonomous missions such as disaster response, infrastructure inspection, and smart city operations. UAVs, leveraging their high mobility, can provide accurate prior maps and wide-area aerial observation for unmanned ground vehicles. However, their long-range perception accuracy is limited. Conversely, UGVs can achieve high-precision environmental perception along their navigation paths using prior maps, but suffer from a constrained field of view. The collaboration between the two platforms complements their respective strengths, thereby enhancing 3D object perception and mapping accuracy in complex scenarios. To address the aforementioned challenges, this study proposes a cross-platform feature fusion method for 3D object perception and an incremental map updating approach for UAVs and UGVs. First, a dynamic SLAM method that integrates an optimized YOLOv8 with ORB-SLAM3 is employed to mitigate map blurring caused by dynamic noise, providing prior map information for UGVs. Second, a multimodal fusion perception model is constructed for UGVs, utilizing attention mechanisms to achieve deep fusion of multimodal Bird’s-Eye-View (BEV) features. This overcomes issues such as diminishing complementarity between modalities and weak temporal feature associations. Finally, an air ground fusion model based on a cross-attention mechanism is developed to fuse aerial view features with ground-based fused BEV features across platforms, yielding a unified feature representation for 3D object detection and generating a fused high-precision map. Experimental results demonstrate that under complex occlusion scenarios in a simulated dataset, the proposed collaborative perception system improves the mean Average Precision (mAP) by 12.7% and 15.7% compared to using a single UAV or a single UGV, respectively, while increasing the map accuracy F1-score by 0.21. This study provides technical support for achieving real-time and accurate air ground collaborative perception in complex dynamic environments.