With the widespread adoption of the Internet of Things (IoT), UAV–vehicle collaborative inspection systems are crucial for large-scale, IoT-enabled monitoring. Empowered by the IoT, these systems optimize resource allocation and boost the efficiency of IoT-based applications. Nevertheless, variable vehicle and UAV speeds due to wind and precipitation complicate path planning and task scheduling in the IoT-integrated setup. To solve this, this study offers an adaptive solution for dynamic, complex-weather scenarios within the IoT framework. A dynamic task-processing model was developed first, using real-time IoT sensor data for better decisions. Then, the KGTSA optimization algorithm was designed. It combines K-means clustering, HGA, and TS, considering UAV and vehicle speed variations in complex weather and making full use of IoT-device data. K-means generates an initial solution, HGA refines it, and TS fine-tunes UAV routes and task assignments. The simulation results show that KGTSA significantly cuts data collection time while maintaining flexibility. It efficiently manages speed and path uncertainties in complex weather, optimizing task efficiency without weather forecasts. Compared to traditional algorithms, KGTSA shortens data collection time and adapts better to dynamic IoT environments for real-world efficiency.
With the deep integration of Internet of Vehicles (IoV) and edge computing technologies, the spatiotemporal dynamics, burstiness, and load fluctuations of user requests pose severe challenges to microservices auto-scaling. Existing static or periodic resource adjustment strategies struggle to adapt to IoV edge environments and often neglect service dependencies and multi-objective optimization synergy, failing to fully utilize implicit regularities like the symmetry in spatiotemporal patterns. This paper proposes a dual-phase dynamic scaling mechanism: for long-term scaling, the Spatio-Temporal Graph Transformer (STGT) is employed to predict traffic flow by capturing correlations in spatial–temporal distributions of vehicle movements, and the improved Multi-objective Graph-based Proximal Policy Optimization (MGPPO) algorithm is applied for proactive resource optimization, balancing trade-offs among conflicting objectives. For short-term bursts, the Fast Load-Aware Auto-Scaling algorithm (FLA) enables rapid instance adjustment based on the M/M/S queuing model, maintaining balanced load distribution across edge nodes—a feature that aligns with the principle of symmetry in system design. The model comprehensively considers request latency, resource consumption, and load balancing, using a multi-objective reward function to guide optimal strategies. Experiments show that STGT significantly improves prediction accuracy, while the combination of MGPPO and FLA reduces request latency and enhances resource utilization stability, validating its effectiveness in dynamic IoV environments.
Autonomous driving perception systems are vulnerable to physical adversarial attacks. Existing defenses largely adopt loosely coupled architectures where visual and kinematic cues are processed in isolation, thus failing to exploit physical spatiotemporal consistency as a structural prior and often struggling to balance adversarial robustness, transferability, accuracy, and efficiency under realistic attacks. We propose a physics-aware trajectory–appearance consistency defense that detects and corrects spatiotemporal inconsistencies by tightly coupling visual semantics with physical dynamics. The module combines a dual-stream spatiotemporal encoder with endogenous feature orchestration and a frequency-domain kinematic embedding, turning tracking artifacts that are usually discarded as noise into discriminative cues. These inconsistencies are quantified by a Trajectory–Appearance Mutual Exclusion (TAME) energy, which supports a physics-aware switching rule to override flawed visual predictions. Operating on detector backbone features, outputs, and tracking states, the defense can be attached as a plug-in module behind diverse object detectors. Experiments on nuScenes, KITTI, and BDD100K show that the proposed defense substantially improves robustness against diverse categories of attacks: on nuScenes, it improves Correction Accuracy (CA) from 86.5% to 92.1% while reducing the computational overhead from 42 ms to 19 ms. Furthermore, the proposed defense maintains over 71.0% CA when transferred to unseen detectors and sustaining 72.4% CA under adaptive attackers.