Exploration Techniques in Reinforcement Learning for Autonomous Vehicles DOI Creative Commons
Ammar Khaleel, Áron Ballagi

Published: Nov. 4, 2024

Autonomous vehicles (AVs) have the potential to revolutionize transportation system by enhancing road safety, reducing traffic congestion, and freeing drivers from monotonous tasks. Effective exploration is essential for AVs navigate safely adapt dynamic environments. Reinforcement learning (RL) enables learn optimal behaviors through continuous interaction with their environment. This paper reviews recent RL research on designing strategies single- multi-agent AV systems. It categorizes methods based underlying principles addresses challenges. analyzes key algorithms' strengths, limitations, empirical performance. By compiling analyzing current state of research, this aims facilitate future advancements in using RL, offering insights into trends directions evolving field.

Language: Английский

A Review of Simultaneous Localization and Mapping Algorithms Based on Lidar DOI Creative Commons
Yong Li, Jianping An, Na He

et al.

World Electric Vehicle Journal, Journal Year: 2025, Volume and Issue: 16(2), P. 56 - 56

Published: Jan. 21, 2025

Simultaneous localization and mapping (SLAM) is one of the key technologies for mobile robots to achieve autonomous driving, lidar SLAM algorithm mainstream research scheme. Firstly, this paper introduces overall framework SLAM, elaborates on functions front-end scan matching, loop closure detection, back-end optimization, map building module, summarizes algorithms used. Then, classical representative are described compared from three aspects: pure algorithm, multi-sensor fusion deep learning algorithm. Finally, challenges faced by in practical use discussed. The development trend prospected five dimensions: lightweight, fusion, combination new sensors, multi-robot collaboration, learning. This can provide a brief guide novices entering field comprehensive reference experienced researchers engineers explore directions.

Language: Английский

Citations

1
An Efficient Three-Dimensional Point Cloud Segmentation Method for the Dimensional Quality Assessment of Precast Concrete Components Utilizing Multiview Information Fusion DOI
Hua‐Ping Wan, Wei Zhang, Yi Chen

et al.

Journal of Computing in Civil Engineering, Journal Year: 2025, Volume and Issue: 39(3)

Published: Feb. 27, 2025

Language: Английский

Citations

1
Comprehensive Performance Evaluation between Visual SLAM and LiDAR SLAM for Mobile Robots: Theories and Experiments DOI Creative Commons
Yu-Lin Zhao, Yi-Tian Hong, Han‐Pang Huang

et al.

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(9), P. 3945 - 3945

Published: May 6, 2024

SLAM (Simultaneous Localization and Mapping), primarily relying on camera or LiDAR (Light Detection Ranging) sensors, plays a crucial role in robotics for localization environmental reconstruction. This paper assesses the performance of two leading methods, namely ORB-SLAM3 SC-LeGO-LOAM, focusing mapping both indoor outdoor environments. The evaluation employs artificial cost-effective datasets incorporating data from 3D an RGB-D (color depth) camera. A practical approach is introduced calculating ground-truth trajectories during benchmarking, reconstruction maps based ground truth are established. To assess performance, ATE RPE utilized to evaluate accuracy localization; standard deviation employed compare stability process different methods. While algorithms exhibit satisfactory positioning accuracy, their suboptimal scenarios with inadequate textures. Furthermore, established by approaches also provided direct observation differences limitations encountered map construction. Moreover, research includes comprehensive comparison computational metrics, encompassing Central Processing Unit (CPU) utilization, memory usage, in-depth analysis. revealed that Visual requires more CPU resources than SLAM, due additional storage requirements, emphasizing impact factors resource requirements. In conclusion, suitable outdoors its nature, while excels indoors, compensating sparse aspects SLAM. facilitate further research, technical guide was researchers related fields.

Language: Английский

Citations

6
RED-SLAM: real-time and effective RGB-D SLAM with spatial-geometric observations and fast semantic perception for dynamic environments DOI
Hailin Liu, Lianfang Tian, Qiliang Du

et al.

Measurement Science and Technology, Journal Year: 2025, Volume and Issue: 36(3), P. 036303 - 036303

Published: Feb. 10, 2025

Abstract Most visual simultaneous localization and mapping (vSLAM) methods assume a static scene, limiting their effectiveness in complex, real-world dynamic environments. This paper presents RED-SLAM-a real-time SLAM method based on the ORB-SLAM3 framework for RGB-D sensors, designed to effectively address impact of objects. RED-SLAM leverages spatial-geometric observations combined with semantic cues identify points within field view, thereby utilizing only state estimation. In geometric verification module, initial distinction between is achieved by checking spatial projection ray distance error matching map feature points. To conserve computational resources, segmentation performed exclusively designated frames, which are constructed changes The detected objects subsequently spread successive frames using propagation technique. All associated excluded further enhance identification accuracy point. Compared existing that apply across all or keyframes, performs when change, improving system’s performance efficiency. Experimental results public datasets scenes demonstrate our enhances pose estimation environments, achieving competitive compared state-of-the-art methods, while maintaining reliable performance.

Language: Английский

Citations

0
Near Real-Time 3D Reconstruction of Construction Sites Based on Surveillance Cameras DOI Creative Commons
Aoran Sun, Xuehui An, Pengfei Li

et al.

Buildings, Journal Year: 2025, Volume and Issue: 15(4), P. 567 - 567

Published: Feb. 12, 2025

The 3D reconstruction of construction sites is great importance for progress, quality, and safety management. Currently, most the existing methods are unable to conduct continuous uninterrupted perception, it difficult achieve registration with real coordinates dimensions. This study proposes a hierarchical framework based on surveillance cameras. method can quickly perform on-site restoration by taking camera images as inputs. It combines 2D features does not need transfer learning or calibration. By experimenting one site, we found that this complete point cloud estimation within an average 3.105 s through images. RMSE site 0.358 m, which better than methods. Through method, data scope cameras be obtained, connection between effectively established. Combined visual information, beneficial digital twin management sites.

Language: Английский

Citations

0
Centerline-based registration for shield tunnel 3D reconstruction using spinning mid-range LiDAR point cloud and multi-cameras DOI
Liao Jian, Wenge Qiu, Yunjian Cheng

et al.

Automation in Construction, Journal Year: 2025, Volume and Issue: 171, P. 105950 - 105950

Published: Jan. 10, 2025

Language: Английский

Citations

0
Cross-area scheduling and conflict-free path planning for multiple robots in non-flat environments DOI
Liwei Yang,

Yun Ge,

Yijiang Zheng

et al.

Expert Systems with Applications, Journal Year: 2025, Volume and Issue: unknown, P. 126767 - 126767

Published: Feb. 1, 2025

Language: Английский

Citations

0
Neural radiance fields for construction site scene representation and progress evaluation with BIM DOI
Yuntae Jeon, Dai Quoc Tran,

Khoa Vo

et al.

Automation in Construction, Journal Year: 2025, Volume and Issue: 172, P. 106013 - 106013

Published: Feb. 8, 2025

Language: Английский

Citations

0
Privacy-Preserved Visual Simultaneous Localization and Mapping Based on a Dual-Component Approach DOI Creative Commons

Mingxu Yang,

Chuhua Huang,

Xin Huang

et al.

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(5), P. 2583 - 2583

Published: Feb. 27, 2025

Edge-assisted visual simultaneous localization and mapping (SLAM) is widely used in autonomous driving, robot navigation, augmented reality for environmental perception, map construction, real-time positioning. However, it poses significant privacy risks, as input images may contain sensitive information, generated 3D point clouds can reconstruct original scenes. To address these concerns, this paper proposes a dual-component privacy-preserving approach SLAM. First, protection method proposed, which combines object detection image inpainting to protect privacy-sensitive information images. Second, an encryption algorithm introduced convert cloud data into line through dimensionality enhancement. Integrated with ORB-SLAM3, the proposed evaluated on Oxford Robotcar KITTI datasets. Results demonstrate that effectively safeguards while ORB-SLAM3 maintains accurate pose estimation dynamic outdoor Furthermore, encrypted prevents unauthorized attacks recovering cloud. This enhances SLAM expected expand its potential applications.

Language: Английский

Citations

0
TSO-HA*-Net: A Hybrid Global Path Planner for the Inspection Vehicles Used in Caged Poultry Houses DOI Creative Commons
Yueping Sun,

Zhanxue Cao,

Wei Yan

et al.

Agriculture, Journal Year: 2025, Volume and Issue: 15(5), P. 532 - 532

Published: Feb. 28, 2025

Traditional track-based inspection schemes for caged poultry houses face issues with vulnerable tracks and cumbersome maintenance, while existing rail-less alternatives lack robust, reliable path planners. This study proposes TSO-HA*-Net, a hybrid global planner that combines TSO-HA* topological planning, which allows the vehicle to continuously traverse predetermined trackless route within each house conduct house-to-house inspections. Initially, spatiotemporally optimized Hybrid A* (TSO-HA*) is employed as lower-level efficiently construct semi-structured network by integrating predefined rules into grid map of houses. Subsequently, Dijkstra’s algorithm adopted plan smooth aligns starting ending poses, conforming network. retains smoothness HA* paths reducing both time computational overhead, thereby enhancing speed efficiency in generation. Experimental results show compared LDP-MAP A*-dis, utilizing distance reference tree (DRT) h2 calculation, total planning reduced 66.6% 96.4%, respectively, stored nodes are 99.7% 97.4%, respectively. The application collision template minimum reduction 4.0% front-end time, prior detection further decreases an average 19.1%. TSO-HA*-Net achieves mere 546.6 ms, addressing critical deficiency viable vehicles provides valuable case studies algorithmic insights similar task.

Language: Английский

Citations

0