Lei Nie

Spatial crowdsourcing (SC) aims to assign spatial tasks to mobile workers while balancing platform utility and matching stability. Existing studies typically focus on either utility maximization or strict stability, while often neglecting heterogeneous constraints such as workers' skills and spatial coverage. In this paper, we study the Approximate Stable Task Assignment with coverage and skill (ASTA) problem, which maximizes system utility under a relaxed stability constraint and is proven to be NP-hard. To address this problem, we propose a two-stage solution framework. In the first stage, we develop a Task-Driven Stability Construction (TDSC) algorithm to generate a feasible initial matching that satisfies stability requirements. In the second stage, we design two optimization algorithms. We first propose a Stability-Guided Revision (SGR) algorithm, which iteratively replaces matched workers with unmatched ones while maintaining stability. To address the limitation in the greedy nature of SGR, we further propose a Multi-Armed Bandit-based Adaptive Large Neighborhood Search (MAB-ALNS) algorithm, which adaptively selects destruction operators based on the Upper Confidence Bound (UCB) policy. We theoretically analyze the regret of the adaptive strategy and derive an upper bound. Finally, extensive experiments on one synthetic dataset and two real-world datasets demonstrate that our methods outperform baseline approaches.

Mobile edge computing (MEC) has emerged as a powerful solution for handling computation-intensive applications at the network edge. Task offloading has become a widely adopted strategy to optimize network resource utilization and deliver enhanced services to user equipment (UE). To meet stringent end-to-end delay and energy consumption requirements, users can offload tasks to edge servers where relevant services are pre-cached. However, the limited storage capacity and the heterogeneity of edge nodes make collaborative cache decision-making and computing resource allocation at the edge critical challenges. We address the optimization of task offloading, resource allocation, service caching, and push decisions within the MEC system through the coordinated efforts of edge servers. The objective is to minimize long-term offloading latency while maximizing the cache hit ratio, presenting a mixed-integer nonlinear programming (MINLP) problem. To tackle this challenge, we propose a novel soft actor-critic approach, transforming the problem into a Markov decision process (MDP). Numerical results demonstrate that our algorithm significantly outperforms existing methods in terms of system cost, average delay, and cache hit ratio.

light detection and ranging (LiDAR) odometry is a critical technology widely applied in pose estimation for autonomous driving and in simultaneous localization and mapping (SLAM). By using a laser scanner, LiDAR captures environmental information to enable precise spatial localization and mapping. However, traditional LiDAR odometry methods mainly depend on static environmental features for positioning and mapping, limiting adaptability in dynamic settings and reducing pose estimation accuracy. To overcome this limitation, we propose ALO, a novel adaptive LiDAR odometry approach designed for dynamic environments. First, an adaptive constant velocity model predicts the expected motion trajectory, supplying prior pose information, while a first-in-first-out voxel grid manages the local map in dynamic conditions. Next, a linear system with dynamic weights based on point-surface residuals is established, minimizing the influence of dynamic features on pose estimation. Finally, the predicted prior pose serves as the initial value for adaptive iterative closest point (ICP) registration, enhancing pose estimation accuracy and enabling real-time local map updates. Extensive experiments on the public KITTI dataset demonstrate that the proposed method achieves at least a 23.69% improvement in pose estimation accuracy over existing solutions.

Simultaneous Localization and Mapping (SLAM) is critical for autonomous driving. Compared with visual SLAM which suffers from variable lighting conditions and sparse features, Light Detection and Ranging (LiDAR) based solution is highly competitive since LiDAR can offer precise distance measurements and robust resistance to changes in illumination. However, current LiDAR SLAM techniques only estimate pose from a single scale, such that complex environments with dynamic objects and similar features can heavily affect the performance, leading to poor accuracy and robustness of pose estimation. To address these challenges, we present MTC-SLAM, a multi-scale and tightly coupled LiDAR-Inertial SLAM method. It begins with a coarse adjustment across six degrees of freedom at a larger scale, followed by adaptive step-size least squares updates for $\boldsymbol{x}$, $\boldsymbol{y}$, and $\boldsymbol{yaw}$ on a smaller scale, and dynamic interpolation updates for $\boldsymbol{z}$, $\boldsymbol{pitch}$, and $\boldsymbol{roll}$. The updated poses are integrated as odometer factors, alongside Global Positioning System (GPS) and loop closure factors, to construct a factor graph for backend optimization. Additionally, a multi-scale loop closure detection method is proposed to identify loops using multi-scale descriptors for global optimization. Extensive experiments on the MulRan dataset demonstrate that MTC-SLAM outperforms existing mainstream methods in both accuracy and robustness.

As a key technology for autonomous navigation and positioning in mobile robots, light detection and ranging (LiDAR) odometry is widely used in autonomous driving applications. The Iterative Closest Point (ICP)-based methods have become the core technique in LiDAR odometry due to their efficient and accurate point cloud registration capability. However, some existing ICP-based methods do not consider the reliability of the initial pose, which may cause the method to converge to a local optimum. Furthermore, the absence of an adaptive mechanism hinders the effective handling of complex dynamic environments, resulting in a significant degradation of registration accuracy. To address these issues, this paper proposes an adaptive ICP-based LiDAR odometry method that relies on a reliable initial pose. First, distributed coarse registration based on density filtering is employed to obtain the initial pose estimation. The reliable initial pose is then selected by comparing it with the motion prediction pose, reducing the initial error between the source and target point clouds. Subsequently, by combining the current and historical errors, the adaptive threshold is dynamically adjusted to accommodate the real-time changes in the dynamic environment. Finally, based on the reliable initial pose and the adaptive threshold, point-to-plane adaptive ICP registration is performed from the current frame to the local map, achieving high-precision alignment of the source and target point clouds. Extensive experiments on the public KITTI dataset demonstrate that the proposed method outperforms existing approaches and significantly enhances the accuracy of LiDAR odometry.

Efficient traffic signal control is a cost-effective way to ease urban traffic congestion. Multi-agent reinforcement learning (MARL) has become a widely adopted method for optimizing traffic signal control (TSC). However, existing MARL-based methods often focus on a single optimization objective, lacking a comprehensive consideration of traffic efficiency, environmental pollution, and traffic safety. Simultaneously, these methods often fail to effectively capture the dynamic and complex interactions among agents in multi-intersection scenarios, which negatively impacts traffic efficiency. In this article, we propose a collaborative MARL-based method for multi-objective TSC, called CMRM. First, we introduce a multi-objective reward mechanism that integrates traffic efficiency, environmental impact, and safety to guide agents toward more comprehensive optimization. Second, we design a cooperation enhancement module (CEM) based on the graph attention mechanism to dynamically capture neighboring agents’ state information. This mitigates the partial observability problem in independent proximal policy optimization (IPPO) and enhances the model’s ability to capture dynamic and complex interactions among agents. Finally, we assess the performance of the proposed CMRM method using SUMO on two real traffic networks. Experimental results demonstrate that our method significantly improves traffic efficiency while reducing environmental pollution and enhancing traffic safety, compared to the best performing baseline, our method reduces CO2 emission by approximately 17.53% and 9.57%, and lowers vehicle collision risks by 44.39% and 42.85% in two different traffic networks.