Uncrewed aerial vehicle (UAV) has emerged as a promising solution for automating railway inspections due to its high mobility, flexible deployment, and reduced labor cost. In this paper, we investigate UAV-assisted railway inspections, which include object recognition, humidity monitoring, and critical infrastructure modeling, each with distinct data volumes and computational requirements. Particularly, we introduce a UAV-assisted railway inspection framework. Different types of sensors are divided into several clusters. The UAV departs from the hive, flies over each cluster to collect their computational requirements, and performs task offloading before returning to the hive. This process is formulated as a joint optimization problem of trajectory planning and task offloading to minimize the weighted sum of latency and energy consumption. Considering the constrained computing and storage capabilities of UAVs, it is crucial but challenging to develop a lightweight yet high-performing solution for the multi-objective optimization problems. As such, a novelArtificial General Intelligence (AGI)-orientedTransformer (AoT) algorithm is proposed to solve the optimization problem. It uses an encoder-only architecture to process either sensor location or task features, and then directs the encoded outputs to different output heads to make decisions on UAV trajectory and task offloading. Simulation results demonstrate that the proposed AoT algorithm outperforms benchmark algorithms in terms of trajectory length and average offloading cost.

4D LiDAR semantic segmentation classifies the semantic category of each LiDAR point and detects whether it is dynamic, a critical ability for tasks like obstacle avoidance and autonomous navigation. Existing approaches often rely on computationally heavy 4D convolutions or recursive networks, which result in poor real-time performance. In this paper, we introduce SegNet4D, a novel real-time 4D semantic segmentation network, offering both efficiency and strong semantic understanding. SegNet4D addresses 4D segmentation as two tasks: single-scan semantic segmentation and moving object segmentation, each tackled by a separate network head. Both results are combined in a motion-semantic fusion module to achieve comprehensive 4D segmentation. Additionally, instance information is extracted from the current scan and exploited for instance-wise segmentation consistency. Extensive experiments on the SemanticKITTI and nuScenes datasets demonstrate that our method outperforms the state-of-the-art in both 4D semantic segmentation and moving object segmentation. Through detailed runtime analysis, our method shows greater efficiency, enabling real-time operation. Besides, its effectiveness and efficiency have also been validated on a real-world robotic platform. The implementation of our method has been released at https: //github.com/nubot-nudt/SegNet4D.

Unmanned Aerial Vehicles (UAVs) provide a flexible working paradigm for device-cloud communication. Besides working as a relay between devices and clouds, UAVs can also provide mobile edge computing (MEC) services. In this paper, we investigate a computation offloading problem for UAV-assisted MEC networks in which local devices, UAVs, and clouds collaboratively process computing tasks to achieve energy-saving and latency reduction. In such green UAV-assisted MEC networks, we treat the same energy consumption of task processing differently due to processing location and assign different weights to the energy consumption of devices, UAVs, and clouds. Specifically, we propose a two-stage computation offloading framework including 1) the device clustering stage to determine the device cluster connected to certain UAVs and 2) the network operation stage to conduct computation offloading. We formulate the offloading process as a stochastic optimization problem to minimize the offloading cost. Furthermore, we decouple the optimization problem into a UAV selection subproblem and an offloading decision subproblem. Particularly, for the former subproblem, we employ a simulated annealing-based algorithm to minimize the total transmit energy of devices and UAVs. For the latter, we utilize a proximal policy optimization-based offloading algorithm to ascertain the processing locations of computing tasks. Simulation results show that the proposed algorithm outperforms in terms of energy reservation and latency reduction.

In this paper, we investigate a resource orchestration and transmission scheduling problem for data-intensive services with diversified service requirements. A three-layer collaborative architecture is presented to support dynamic networking and computing resource allocation. To obtain optimal orchestration and scheduling policies, we formulate a constrained resource scheduling problem with the objective to maximizing resource utilization and scheduling success ratio. Since the complicated coupled constraints among decisions, we decouple the problem into a two-stage sub-problems of resource orchestration and transmission scheduling. To realize cross-domain resource orchestration and deterministic transmission of large-scale computing tasks, a two-stage resource scheduling scheme is proposed. Specifically, the first stage makes the resource orchestration decision by a greedy algorithm, and the second stage makes the transmission scheduling decision based on a deep reinforcement learning algorithm. Simulation results show that the proposed solution can effectively improve resource utilization and scheduling success ratio while satisfying diversified service requirements, as compared with benchmarks.

Integration of the fifth-generation mobile communication technology (5G) into time-sensitive networking (TSN) was first proposed in the 3GPP Release 16. However, this conceptual proposal lacks of detailed designs to guarantee bounded latency and high reliability of this integration. In this paper, we study a deterministic scheduling problem for TSN-5G integrated networks in industrial Internet of things (IIoT) scenarios, in which a unified control plane jointly allocates the time-frequency resources for TSN and 5G to support deterministic end-to-end transmission. Specifically, we design a novel control architecture, i.e., centralized network and distributed user, for the integrated networks to reduce the signaling overhead. Moreover, we formulate a stochastic optimization problem for IIoT scenarios to maximize the number of successfully scheduled flows as well as realize throughput fairness for wired and wireless equipment. Since the resource allocation of TSN and 5G are coupled, this problem is NP-hard. We propose a dueling double deep Q network (D3QN) based Joint Resource Allocation (DJRA) algorithm. By leveraging two convolution-enhanced neural networks, with their parameters periodically synchronized, the accuracy of the estimated Q-value can be increased and the convergence speed of DJRA can be accelerated. Simulation results show that the proposed algorithm can facilitate efficient cooperation between TSN and 5G as compared to the other heuristic and learning-based algorithms.

Identifying moving objects is a crucial capability for autonomous navigation, consistent map generation, and future trajectory prediction of objects. In this paper, we propose a novel network that addresses the challenge of segmenting moving objects in 3D LiDAR scans. Our approach not only predicts point-wise moving labels but also detects instance information of main traffic participants. Such a design helps determine which instances are actually moving and which ones are temporarily static in the current scene. Our method exploits a sequence of point clouds as input and quantifies them into 4D voxels. We use 4D sparse convolutions to extract motion features from the 4D voxels and inject them into the current scan. Then, we extract spatio-temporal features from the current scan for instance detection and feature fusion. Finally, we design an upsample fusion module to output point-wise labels by fusing the spatio-temporal features and predicted instance information. We evaluated our approach on the LiDAR-MOS benchmark based on SemanticKITTI and achieved better moving object segmentation performance compared to state-of-the-art methods, demonstrating the effectiveness of our approach in integrating instance information for moving object segmentation. Furthermore, our method shows superior performance on the Apollo dataset with a pre-trained model on SemanticKITTI, indicating that our method generalizes well in different scenes. The code and pre-trained models of our method will be released at https://github.com/nubot-nudt/InsMOS.