Path planning algorithms in the autonomous driving system: A comprehensive review

Robotics and Autonomous Systems, Год журнала: 2024, Номер 174, С. 104630 - 104630

Опубликована: Янв. 13, 2024

This comprehensive review focuses on the Autonomous Driving System (ADS), which aims to reduce human errors that are reason for about 95% of car accidents. The ADS consists six stages: sensors, perception, localization, assessment, path planning, and control. We explain main state-of-the-art techniques used in each stage, analyzing 275 papers, with 162 specifically planning due its complexity, NP-hard optimization nature, pivotal role ADS. paper categorizes into three primary groups: traditional (graph-based, sampling-based, gradient-based, optimization-based, interpolation curve algorithms), machine deep learning, meta-heuristic optimization, detailing their advantages drawbacks. Findings show methods, representing 23% our study, preferred being general problem solvers capable handling complex problems. In addition, they have faster convergence reduced risk local minima. Machine learning techniques, accounting 25%, favored capabilities fast responses known scenarios. trend toward hybrid algorithms (27%) combines various merging algorithm's benefits overcoming other's Moreover, adaptive parameter tuning is crucial enhance efficiency, applicability, balancing search capability. sheds light future autonomous driving systems, helping tackle current challenges unlock full vehicles.

Язык: Английский

---

A Comprehensive Review of Deep Learning: Architectures, Recent Advances, and Applications

Information, Год журнала: 2024, Номер 15(12), С. 755 - 755

Опубликована: Ноя. 27, 2024

Deep learning (DL) has become a core component of modern artificial intelligence (AI), driving significant advancements across diverse fields by facilitating the analysis complex systems, from protein folding in biology to molecular discovery chemistry and particle interactions physics. However, field deep is constantly evolving, with recent innovations both architectures applications. Therefore, this paper provides comprehensive review DL advances, covering evolution applications foundational models like convolutional neural networks (CNNs) Recurrent Neural Networks (RNNs), as well such transformers, generative adversarial (GANs), capsule networks, graph (GNNs). Additionally, discusses novel training techniques, including self-supervised learning, federated reinforcement which further enhance capabilities models. By synthesizing developments identifying current challenges, insights into state art future directions research, offering valuable guidance for researchers industry experts.

Язык: Английский

---

Survey of Autonomous Vehicles’ Collision Avoidance Algorithms

Sensors, Год журнала: 2025, Номер 25(2), С. 395 - 395

Опубликована: Янв. 10, 2025

Since the field of autonomous vehicles is developing quickly, it becoming increasingly crucial for them to safely and effectively navigate their surroundings avoid collisions. The primary collision avoidance algorithms currently employed by self-driving cars are examined in this thorough survey. It looks into several methods, such as sensor-based methods precise obstacle identification, sophisticated path-planning that guarantee follow dependable safe paths, decision-making systems allow adaptable reactions a range driving situations. survey also emphasizes how Machine Learning can improve efficacy avoidance. Combined, these techniques necessary enhancing dependability safety systems, ultimately increasing public confidence game-changing technology.

Язык: Английский

---

Trajectory planning and tracking control in autonomous driving system: Leveraging machine learning and advanced control algorithms

Engineering Science and Technology an International Journal, Год журнала: 2025, Номер 64, С. 101950 - 101950

Опубликована: Фев. 18, 2025

Язык: Английский

---

Deep learning for autonomous driving systems: technological innovations, strategic implementations, and business implications - a comprehensive review

Complex Engineering Systems, Год журнала: 2025, Номер 5(1)

Опубликована: Фев. 12, 2025

The rapid advancements in deep learning have significantly transformed the landscape of autonomous driving, with profound technological, strategic, and business implications. Autonomous driving systems, which rely on to enhance real-time perception, decision-making, control, are poised revolutionize transportation by improving safety, efficiency, mobility. Despite this progress, numerous challenges remain, such as data processing, decision-making under uncertainty, navigating complex environments. This comprehensive review explores state-of-the-art methodologies, including Convolutional Neural Networks (CNNs), Recurrent Networks, Long Short-Term Memory networks, transformers that central tasks object detection, scene understanding, path planning. Additionally, examines strategic implementations, focusing integration into automotive sector, scalability artificial intelligence-driven their alignment regulatory safety standards. Furthermore, study highlights implications adoption, its influence operational competitive dynamics, workforce requirements. literature also identifies gaps, particularly achieving full autonomy (Level 5), sensor fusion, addressing long-term costs challenges. By these issues, has potential redefine future mobility, enabling safer, more efficient, fully systems. aims provide insights for stakeholders, manufacturers, intelligence developers, policymakers, navigate complexities integrating driving.

Язык: Английский

---

A visual detection algorithm for autonomous driving road environment perception

Engineering Applications of Artificial Intelligence, Год журнала: 2024, Номер 133, С. 108034 - 108034

Опубликована: Фев. 7, 2024

Язык: Английский

---

Vision-Based On-Road Nighttime Vehicle Detection and Tracking Using Improved HOG Features

Sensors, Год журнала: 2024, Номер 24(5), С. 1590 - 1590

Опубликована: Фев. 29, 2024

The lack of discernible vehicle contour features in low-light conditions poses a formidable challenge for nighttime detection under hardware cost constraints. Addressing this issue, an enhanced histogram oriented gradients (HOGs) approach is introduced to extract relevant features. Initially, lights are extracted using combination background illumination removal and saliency model. Subsequently, these integrated with template-based delineate regions containing potential vehicles. In the next step, fusion superpixel HOG (S-HOG) within performed, support vector machine (SVM) employed classification. A non-maximum suppression (NMS) method applied eliminate overlapping areas, incorporating vertical histograms symmetrical (V-HOGs). Finally, Kalman filter utilized tracking candidate vehicles over time. Experimental results demonstrate significant improvement accuracy recognition scenarios proposed method.

Язык: Английский

---

Multitask Knowledge Distillation Guides End-to-End Lane Detection

IEEE Transactions on Industrial Informatics, Год журнала: 2023, Номер 19(9), С. 9703 - 9712

Опубликована: Янв. 3, 2023

Autonomous driving has witnessed rapid development with the application of artificial intelligence technology in recent years. Lane detection is one tasks environment perception, which affects planning and decision-making directly, requires algorithm to meet both high precision efficiency. Most existing methods extract pixels belonging lanes image, should be postprocessed, otherwise it cannot applied subsequent like planning. This article proposes an end-to-end lane method that utilizes auxiliary supervision knowledge distillation based teaching-test module predict parameters polynomials directly. The guides polynomial regression branch learn shape features from segmentation improve fitting accuracy under complex road conditions. proposed validated on TuSimple CULane datasets, competitive state-of-the-art efficiency accuracy.

Язык: Английский

---

LLDNet: A Lightweight Lane Detection Approach for Autonomous Cars Using Deep Learning

Sensors, Год журнала: 2022, Номер 22(15), С. 5595 - 5595

Опубликована: Июль 26, 2022

Lane detection plays a vital role in making the idea of autonomous car reality. Traditional lane methods need extensive hand-crafted features and post-processing techniques, which make models specific feature-oriented, susceptible to instability for variations on road scenes. In recent years, Deep Learning (DL) models, especially Convolutional Neural Network (CNN) have been proposed utilized perform pixel-level segmentation. However, most focus achieving high accuracy while considering structured roads good weather conditions do not put emphasis testing their defected roads, ones with blurry lines, no cracked pavements, are predominant real world. Moreover, many these CNN-based complex structures require high-end systems operate, makes them quite unsuitable being implemented embedded devices. Considering shortcomings, this paper, we introduced novel CNN model named LLDNet based an encoder–decoder architecture that is lightweight has tested adverse as well conditions. A channel attention spatial module integrated into designed refine feature maps outstanding results lower number parameters. We used hybrid dataset train our model, was created by combining two separate datasets, compared few state-of-the-art architectures. Numerical show surpasses terms dice coefficient, IoU, size models. carried out experiments videos different Bangladesh. The visualization exhibit can detect lanes accurately both Experimental elicit method capable detecting ready practical implementation.

Язык: Английский

---

Deep-Learning-Based Network for Lane Following in Autonomous Vehicles

Electronics, Год журнала: 2022, Номер 11(19), С. 3084 - 3084

Опубликована: Сен. 27, 2022

The research field of autonomous self-driving vehicles has recently become increasingly popular. In addition, motion-planning technology is essential for because it mitigates the prevailing on-road obstacles. Herein, a deep-learning-network-based architecture that was integrated with VGG16 and gated recurrent unit (GRU) applied lane-following on roads. normalized input image fed to three-layer output layer as pattern GRU last layer. Next, processed data were two fully connected layers, dropout added in between each Afterward, evaluate model, steering angle speed from control task predicted parameters. Experiments conducted using dataset Udacity simulator real dataset. results show proposed framework remarkably angles different directions. Furthermore, approach achieved higher mean square errors 0.0230 0.0936 inference times 3–4 3 ms. We also implemented our NVIDIA Jetson embedded platform (Jetson Nano 4 GB) compared GPU’s computational time. revealed system took 45–46 s execute single epoch order predict angle. generates fruitful accurate motion planning driving.

Язык: Английский

---