Artificial Intelligence in Navigation

Topic Information

Dear Colleagues,

Navigation is the science and technology of accurately determining the position and velocity of an airborne, land, or marine vehicle relative to a known reference, wherein the planning and execution of the maneuvers necessary to move between desired locations are analyzed. Artificial intelligence (AI) is a wide-ranging branch of computer science concerning the design of machines and software capable of performing tasks associated with intelligent beings, which encompasses machine learning, expert systems, knowledge engineering, and language processing. Due to the accelerating development and widespread prevalence of AI, navigation can effectively be supported by newly established AI technologies. This Topic is open to the entire spectrum of basic and applied scientific research, novel practical applications, and innovations. It also promotes the dialogue between governmental and private actors or institutions. Contributions to navigation theory, systems design, interoperability, or social impact analyses are similarly welcome.

Topics include, but are not limited to, the following:

• Marine, land, and airborne navigation
• Positioning/localization, route planning, and guidance
• Vehicle control
• VR/AR technologies, computer vision, and image processing/analysis/understanding
• Voice/speech recognition, language processing, and machine translation
• Methodologies, frameworks, and models in artificial intelligence
• Machine learning, deep learning, neural networks, fuzzy systems, evolutionary computation, and agent-based systems
• Knowledge representation and reasoning
• Information systems, maps, databases, and open data sources
• Robotic systems and autonomous vehicles
• Commercial, military, intelligence, education, and research applications
• Security and privacy issues and ethical questions
• Best practices and use cases.

Prof. Dr. Arpad Barsi
Prof. Dr. Eliseo Clementini
Topic Editors

Keywords

Participating Journals

Journal Name	Impact Factor	CiteScore	Launched Year	First Decision (median)	APC
AI ai	3.1	7.2	2020	17.6 Days	CHF 1600
Applied Sciences applsci	2.5	5.3	2011	17.8 Days	CHF 2400
Electronics electronics	2.6	5.3	2012	16.8 Days	CHF 2400
ISPRS International Journal of Geo-Information ijgi	2.8	6.9	2012	36.2 Days	CHF 1700
Remote Sensing remotesensing	4.2	8.3	2009	24.7 Days	CHF 2700
Robotics robotics	2.9	6.7	2012	17.7 Days	CHF 1800
Sensors sensors	3.4	7.3	2001	16.8 Days	CHF 2600

MDPI Topics is cooperating with Preprints.org and has built a direct connection between MDPI journals and Preprints.org. Authors are encouraged to enjoy the benefits by posting a preprint at Preprints.org prior to publication:

1. Immediately share your ideas ahead of publication and establish your research priority;
2. Protect your idea from being stolen with this time-stamped preprint article;
3. Enhance the exposure and impact of your research;
4. Receive feedback from your peers in advance;
5. Have it indexed in Web of Science (Preprint Citation Index), Google Scholar, Crossref, SHARE, PrePubMed, Scilit and Europe PMC.

Published Papers (18 papers)

Order results

Content type Publication Date

Result details

Normal Extended Compact

Journals

All AI Applied Sciences Electronics IJGI Remote Sensing Robotics Sensors

Show export options expand_more Show export options expand_less

Select all

Export citation of selected articles as:

Plain Text BibTeX BibTeX (without abstracts) Endnote Endnote (without abstracts) Tab-delimited RIS

Error

Oops... you haven't selected anything for export.

Ok

clear

23 pages, 8508 KiB  

Open AccessArticle

An Evaluation of Optimization Algorithms for the Optimal Selection of GNSS Satellite Subsets

by Abdulaziz Alluhaybi, Panos Psimoulis and Rasa Remenyte-Prescott

Remote Sens. 2024, 16(10), 1794; https://doi.org/10.3390/rs16101794 - 18 May 2024

Viewed by 978

Abstract

Continuous advancements in GNSS systems have led, apart from the broadly used GPS, to the development of other satellite systems (Galileo, BeiDou, GLONASS), which have significantly increased the number of available satellites for GNSS positioning applications. However, despite GNSS satellites’ redundancy, a potential [...] Read more.

Continuous advancements in GNSS systems have led, apart from the broadly used GPS, to the development of other satellite systems (Galileo, BeiDou, GLONASS), which have significantly increased the number of available satellites for GNSS positioning applications. However, despite GNSS satellites’ redundancy, a potential poor GNSS satellite signal (i.e., low signal-to-noise ratio) can negatively affect the GNSS’s performance and positioning accuracy. On the other hand, selecting high-quality GNSS satellite signals by retaining a sufficient number of GNSS satellites can enhance the GNSS’s positioning performance. Various methods, including optimization algorithms, which are also commonly adopted in artificial intelligence (AI) methods, have been applied for satellite selection. In this study, five optimization algorithms were investigated and assessed in terms of their ability to determine the optimal GNSS satellite constellation, such as Artificial Bee Colony optimization (ABC), Ant Colony Optimization (ACO), Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Simulated Annealing (SA). The assessment of the optimization algorithms was based on two criteria, such as the robustness of the solution for the optimal satellite constellation and the time required to find the solution. The selection of the GNSS satellites was based on the weighted geometric dilution of precision (WGDOP) parameter, where the geometric dilution of precision (GDOP) is modified by applying weights based on the quality of the satellites’ signal. The optimization algorithms were tested on the basis of 24 h of tracking data gathered from a permanent GNSS station, for GPS-only and multi-GNSS data (GPS, GLONASS, and Galileo). According to the comparison results, the ABC, ACO, and PSO algorithms were equivalent in terms of selection accuracy and speed. However, ABC was determined to be the most suitable algorithm due it requiring the fewest number of parameters to be set. To further investigate ABC’s performance, the method was applied for the selection of an optimal GNSS satellite subset according to the number of total available tracked GNSS satellites (up to 31 satellites), leading to more than 300 million possible combinations of 15 GNSS satellites. ABC was able to select the optimal satellite subsets with 100% accuracy. Full article

(This article belongs to the Topic Artificial Intelligence in Navigation)

►▼ Show Figures

Figure 1

Figure 1
<p>Representation of the ABC searching process and the roles of employed scout and onlooker bees.</p> Full article ">Figure 2
<p>Schematic representation of the solution building process by ants in ACO.</p> Full article ">Figure 3
<p>Representation of the GA processing steps.</p> Full article ">Figure 4
<p>Representation of PSO travelling technique for the solution optimization.</p> Full article ">Figure 5
<p>Schematic representation of SA algorithm procedure.</p> Full article ">Figure 6
<p>Time period of satellites’ mean movement by one degree, considering satellite azimuth and elevation angles.</p> Full article ">Figure 7
<p>(<b>Left</b>) View of the roof of NGI building, with the location of control point NGB2 and (<b>right</b>) the GNSS antenna installed on the top of the pillar of NGB2.</p> Full article ">Figure 8
<p>Number of available GNSS satellites at NGB2 GNSS station for a 24 h period on 20 September 2021.</p> Full article ">Figure 9
<p>Sample of the file of the GPS data information, which includes (i) date–time, (ii) satellite PRN, (iii) azimuth, (iv) elevation angle, and (v) CNR (in dB-Hz).</p> Full article ">Figure 10
<p>The possible combinations of satellite constellations for (<b>left</b>) GPS-only in the cases of 8 and 13 available GPS satellites and (<b>right</b>) multi-GNSS satellite constellation in the case of 18 and 31 available GNSS satellites.</p> Full article ">Figure 11
<p>(<b>Left</b>) The quality of match (accuracy) of the selection of the optimal GPS satellite subset by the optimization algorithms with respect to the actual optimal GPS satellite subset derived by the TM. (<b>Right</b>) The time required for the TM and the optimization algorithms to perform optimal GPS satellite subset selection.</p> Full article ">Figure 12
<p>The comparison of the performance of the optimization algorithms with respect to TM, expressed as the difference between the CNR-WGDOP of the optimal satellite constellation of each optimization algorithm and the corresponding CNR-WDGOP of TM. The results of the four cases of GPS satellite constellations (4, 5, 6, and 7 satellites) are presented. On the left axis, the CNR-WGDOP value of the optimal satellite constellation based on the TM is presented, and on the right axis is the difference between each of the optimization algorithms and the TM.</p> Full article ">Figure 13
<p>The sky plots of epoch 55 (<b>left</b>) and epoch 184 (<b>right</b>) presenting the selection of the optimal GPS satellite subset by the ACO and the TM, and showing the satellites commonly selected (blue) by the two methods, but also those that were differently selected by ACO (yellow) and TM (orange).</p> Full article ">Figure 14
<p>Same as <a href="#remotesensing-16-01794-f013" class="html-fig">Figure 13</a>, this figure presents the sky plots for epoch 81 (<b>left</b>) and epoch 215 (<b>right</b>), as well as differences between the selection of the optimal GPS satellite subset for PSO and TM.</p> Full article ">Figure 15
<p>(<b>Left</b>) The accuracy of the selection of the optimal GNSS satellite subset of ABC with respect to TM for the various cases of satellite constellations and parameter settings and (<b>right</b>) the required time of the TM and ABC algorithm to compute the selection of optimal GNSS satellite subset.</p> Full article ">Figure 16
<p>A comparison of the performance of the ABC algorithm for the three sets of parameter settings with respect to TM, expressed as the difference between the CNR-WGDOP of the optimal satellite constellation of each ABC parameter setting and the corresponding CNR-WDGOP of TM. On the left axis, the CNR-WGDOP value of the optimal satellite constellation based on the TM is presented, and on the right axis is the difference between each of the ABC parameter settings and the TM.</p> Full article ">Figure 17
<p>Sky plots of epochs 20 (<b>left</b>) and 88 (<b>right</b>), presenting the difference in the selection of optimal GNSS satellite subset between ABC setting 1 and the actual TM, by showing the common satellites (blue) and the differences between ABC’s (yellow) and TM’s (orange) satellite selection.</p> Full article ">

20 pages, 4786 KiB  

Open AccessArticle

VIS-SLAM: A Real-Time Dynamic SLAM Algorithm Based on the Fusion of Visual, Inertial, and Semantic Information

by Yinglong Wang, Xiaoxiong Liu, Minkun Zhao and Xinlong Xu

ISPRS Int. J. Geo-Inf. 2024, 13(5), 163; https://doi.org/10.3390/ijgi13050163 - 13 May 2024

Cited by 2 | Viewed by 1169

Abstract

A deep learning-based Visual Inertial SLAM technique is proposed in this paper to ensure accurate autonomous localization of mobile robots in environments with dynamic objects. Addressing the limitations of real-time performance in deep learning algorithms and the poor robustness of pure visual geometry [...] Read more.

A deep learning-based Visual Inertial SLAM technique is proposed in this paper to ensure accurate autonomous localization of mobile robots in environments with dynamic objects. Addressing the limitations of real-time performance in deep learning algorithms and the poor robustness of pure visual geometry algorithms, this paper presents a deep learning-based Visual Inertial SLAM technique. Firstly, a non-blocking model is designed to extract semantic information from images. Then, a motion probability hierarchy model is proposed to obtain prior motion probabilities of feature points. For image frames without semantic information, a motion probability propagation model is designed to determine the prior motion probabilities of feature points. Furthermore, considering that the output of inertial measurements is unaffected by dynamic objects, this paper integrates inertial measurement information to improve the estimation accuracy of feature point motion probabilities. An adaptive threshold-based motion probability estimation method is proposed, and finally, the positioning accuracy is enhanced by eliminating feature points with excessively high motion probabilities. Experimental results demonstrate that the proposed algorithm achieves accurate localization in dynamic environments while maintaining real-time performance. Full article

(This article belongs to the Topic Artificial Intelligence in Navigation)

►▼ Show Figures

Figure 1

Figure 1
<p>System framework diagram.</p> Full article ">Figure 2
<p>Non-blocking semantic information extraction diagram.</p> Full article ">Figure 3
<p>Motion probability grading model.</p> Full article ">Figure 4
<p>The reprojection error model for dynamic features. (<b>a</b>) Moving in a non-epipolar direction; (<b>b</b>) moving in an epipolar direction.</p> Full article ">Figure 5
<p>Shows the comparison of feature selection results for different algorithms on test image 1. (<b>a</b>) The coarse segmentation algorithm; (<b>b</b>) the corrected results.</p> Full article ">Figure 6
<p>Shows the comparison of feature selection results for different algorithms on test image 2. (<b>a</b>) The coarse segmentation algorithm; (<b>b</b>) the corrected results.</p> Full article ">Figure 7
<p>Results of the (<b>a</b>) fr3/walking/xyz, (<b>b</b>) fr3/walking/halfsphere, (<b>c</b>) fr3/walking/rpy, and (<b>d</b>) fr3/walking/static sequences for the TUM dynamic RGB-D datasets. We compared the trajectories estimated by our methods and by the original ORB-SLAM3 with the ground truth.</p> Full article ">Figure 8
<p>Comparison of localization trajectory results for (<b>a</b>) ORB-SLAM3, (<b>b</b>) CS-SLAM, and (<b>c</b>) VIS-SLAM.</p> Full article ">Figure 9
<p>Error comparison of different algorithms in different directions. (<b>a</b>) x-direction; (<b>b</b>) y-direction; (<b>c</b>) z-direction.</p> Full article ">Figure 10
<p>Partial scenes from the experimental environment.</p> Full article ">Figure 11
<p>Diagram of experimental hardware platform and software framework. (<b>a</b>) Hardware device; (<b>b</b>) software framework.</p> Full article ">Figure 12
<p>Comparison of localization results between the two algorithms. (<b>a</b>) ORB-SLAM3; (<b>b</b>) VIS-SLAM.</p> Full article ">Figure 13
<p>Error comparison of different algorithms in different directions. (<b>a</b>) x-direction; (<b>b</b>) y-direction; (<b>c</b>) z-direction.</p> Full article ">

19 pages, 899 KiB  

Open AccessArticle

5G Indoor Positioning Error Correction Based on 5G-PECNN

by Shan Yang, Qiyuan Zhang, Longxing Hu, Haina Ye, Xiaobo Wang, Ti Wang and Syuan Liu

Sensors 2024, 24(6), 1949; https://doi.org/10.3390/s24061949 - 19 Mar 2024

Viewed by 1244

Abstract

With the development of the mobile network communication industry, 5G has been widely used in the consumer market, and the application of 5G technology for indoor positioning has emerged. Like most indoor positioning techniques, the propagation of 5G signals in indoor spaces is [...] Read more.

With the development of the mobile network communication industry, 5G has been widely used in the consumer market, and the application of 5G technology for indoor positioning has emerged. Like most indoor positioning techniques, the propagation of 5G signals in indoor spaces is affected by noise, multipath propagation interference, installation errors, and other factors, leading to errors in 5G indoor positioning. This paper aims to address these issues by first constructing a 5G indoor positioning dataset and analyzing the characteristics of 5G positioning errors. Subsequently, we propose a 5G Positioning Error Correction Neural Network (5G-PECNN) based on neural networks. This network employs a multi-level fusion network structure designed to adapt to the error characteristics of 5G through adaptive gradient descent. Experimental validation demonstrates that the algorithm proposed in this paper achieves superior error correction within the error region, significantly outperforming traditional neural networks. Full article

(This article belongs to the Topic Artificial Intelligence in Navigation)

►▼ Show Figures

Figure 1

Figure 1
<p>Schematic diagram of AOA localization, where the black line is the normal signal propagation, the green dashed line is the noise interference schematic, and the orange dashed line is the multipath influence schematic.</p> Full article ">Figure 2
<p>Schematic plan of the plant workspace; the blue point is the 5G base station. All areas are within the factory’s working space. The light-blue area represents the open space inside the factory. The blue rectangle below indicates operational equipment, with an aisle in the middle for pedestrian passage.</p> Full article ">Figure 3
<p>Contour plot of the error of the measurement results at various locations in the indoor work area. The numbers on the contour lines represent positioning errors, and areas with the same color indicate the same range of positioning errors.</p> Full article ">Figure 4
<p>Schematic diagram of partitions and bar chart of average errors. (<b>a</b>) Schematic diagram of work area zones. The regions labeled A–E represent five distinct areas delineated in the study. (<b>b</b>) Bar chart of average errors in each region.</p> Full article ">Figure 5
<p>The schematic diagram of the constructed neural network model A. The three connected blue rectangles represent the hidden layers of the network. In the Feature structure, the numbers of neurons in the hidden layers are 16, 8, and 4, respectively, while in the Correction structure, they are 30, 16, and 8, respectively.</p> Full article ">Figure 6
<p>The schematic diagram of the constructed neural network model B. The three connected blue rectangles represent the hidden layers of the network. In the Feature structure, the numbers of neurons in the hidden layers are 16, 8, and 4, respectively, while in the Correction structure, they are 6, 8, and 16, respectively.</p> Full article ">Figure 7
<p>Algorithm Flowchart of the 5G Positioning Error Correction Neural Network based on an Adaptive Global Gradient Descent Algorithm.</p> Full article ">Figure 8
<p>Overall test effect diagram. (<b>a</b>) Original error distribution. (<b>b</b>) Corrected error distribution.</p> Full article ">Figure 9
<p>Performance comparison chart for different models in different regions.</p> Full article ">

29 pages, 33317 KiB  

Open AccessArticle

Method for the Identification and Classification of Zones with Vehicular Congestion

by Gary Reyes, Roberto Tolozano-Benites, Laura Lanzarini, César Estrebou, Aurelio F. Bariviera and Julio Barzola-Monteses

ISPRS Int. J. Geo-Inf. 2024, 13(3), 73; https://doi.org/10.3390/ijgi13030073 - 28 Feb 2024

Viewed by 1679

Abstract

Persistently, urban regions grapple with the ongoing challenge of vehicular traffic, a predicament fueled by the incessant expansion of the population and the rise in the number of vehicles on the roads. The recurring challenge of vehicular congestion casts a negative influence on [...] Read more.

Persistently, urban regions grapple with the ongoing challenge of vehicular traffic, a predicament fueled by the incessant expansion of the population and the rise in the number of vehicles on the roads. The recurring challenge of vehicular congestion casts a negative influence on urban mobility, thereby diminishing the overall quality of life of residents. It is hypothesized that a dynamic clustering method of vehicle trajectory data can provide an accurate and up-to-date representation of real-time traffic behavior. To evaluate this hypothesis, data were collected from three different cities: San Francisco, Rome, and Guayaquil. A dynamic clustering algorithm was applied to identify traffic congestion patterns, and an indicator was applied to identify and evaluate the congestion conditions of the areas. The findings indicate a heightened level of precision and recall in congestion classification when contrasted with an approach relying on static cells. Full article

(This article belongs to the Topic Artificial Intelligence in Navigation)

►▼ Show Figures

Figure 1

Figure 1
<p>Components of the proposed method.</p> Full article ">Figure 2
<p>Elements that compose a cluster.</p> Full article ">Figure 3
<p>Cluster-related information.</p> Full article ">Figure 4
<p>Percentage of forgetting and relevance of data per unit of elapsed time for a lambda value set at 0.05.</p> Full article ">Figure 5
<p>Clusters projected on the map.</p> Full article ">Figure 6
<p>Cluster projected on the grid.</p> Full article ">Figure 7
<p>Dynamism of a cluster as a function of elapsed time: 1 s (<b>a</b>), 2 s (<b>b</b>), 3 s (<b>c</b>) and 4 s (<b>d</b>).</p> Full article ">Figure 8
<p>Processed data flow in function of time: 1 s (<b>a</b>), 2 s (<b>b</b>), 3 s (<b>c</b>) and 4 s (<b>d</b>).</p> Full article ">Figure 9
<p>Static grid showing the distribution of the cells that make it up.</p> Full article ">Figure 10
<p>Selection of dynamic road segments at the snapshots of minutes 1 (<b>a</b>), 3 (<b>b</b>), and 5 (<b>c</b>).</p> Full article ">Figure 11
<p>Choosing road segments to establish fixed cells.</p> Full article ">Figure 12
<p>Example of different network complexities: (<b>a</b>) a single road, (<b>b</b>) four roads, and (<b>c</b>) eleven roads.</p> Full article ">

17 pages, 7209 KiB  

Open AccessArticle

MixedSCNet: LiDAR-Based Place Recognition Using Multi-Channel Scan Context Neural Network

by Yan Si, Wenyi Han, Die Yu, Baizhong Bao, Jian Duan, Xiaobin Zhan and Tielin Shi

Electronics 2024, 13(2), 406; https://doi.org/10.3390/electronics13020406 - 18 Jan 2024

Viewed by 1278

Abstract

In the realm of LiDAR-based place recognition tasks, three predominant methodologies have emerged: manually crafted feature descriptor-based methods, deep learning-based methods, and hybrid methods that combine the former two. Manually crafted feature descriptors often falter in reverse visits and confined indoor environments, while [...] Read more.

In the realm of LiDAR-based place recognition tasks, three predominant methodologies have emerged: manually crafted feature descriptor-based methods, deep learning-based methods, and hybrid methods that combine the former two. Manually crafted feature descriptors often falter in reverse visits and confined indoor environments, while deep learning-based methods exhibit limitations in terms of generalization to distinct data domains. Hybrid methods tend to fix these problems, albeit at the cost of an expensive computational burden. In response to this, this paper introduces MixedSCNet, a novel hybrid approach designed to harness the strengths of manually crafted feature descriptors and deep learning models while keeping a relatively low computing overhead. MixedSCNet starts with constructing a BEV descriptor called MixedSC, which takes height, intensity, and smoothness into consideration simultaneously, thus offering a more comprehensive representation of the point cloud. Subsequently, MixedSC is fed into a compact Convolutional Neural Network (CNN), which further extracts high-level features, ultimately yielding a discriminative global point cloud descriptor. This descriptor is then employed for place retrieval, effectively bridging the gap between manually crafted feature descriptors and deep learning models. To substantiate the efficacy of this amalgamation, we undertake an extensive array of experiments on the KITTI and NCLT datasets. Results show that MixedSCNet stands out as the sole method showcasing state-of-the-art performance across both datasets, outperforming the other five methods while maintaining a relatively short runtime. Full article

(This article belongs to the Topic Artificial Intelligence in Navigation)

►▼ Show Figures

Figure 1

Figure 1
<p>System framework of the proposed method.</p> Full article ">Figure 2
<p>Visualization of MixedSC. The light-blue region corresponds to a ring; the pale-yellow area represents a sector; and the light-green area indicates the bin resulting from their intersection. The amalgamation of the three right-side matrices results in a MixedSC descriptor.</p> Full article ">Figure 3
<p>The network architecture of MixedSCNet. <span class="html-italic">Block</span> denotes the number of convolutional operations. 1 × 1, 3 × 3, and 5 × 5 represent kernel sizes.</p> Full article ">Figure 4
<p>Intensity and smoothness channel visualization for a reverse visit case. (<b>a</b>,<b>b</b>) Reverse visit in NCLT 2012-02-02 sequence (<b>c</b>,<b>d</b>) and corresponding intensity channels produced by Intensity Scan Context, where the latter one is the 180°-shifted version of the source channel. (<b>e</b>,<b>f</b>) The smoothness channels of their corresponding MixedSC.</p> Full article ">Figure 5
<p>Feature map visualization for MixedSCNet. The feature vector generated by MixedSCNet is reshaped into a 32 × 32 matrix, which stands for the final feature map.</p> Full article ">Figure 6
<p>Recall–candidate number curve for all sequences.</p> Full article ">

24 pages, 21652 KiB  

Open AccessArticle

A Multi-Source-Data-Assisted AUV for Path Cruising: An Energy-Efficient DDPG Approach

by Tianyu Xing, Xiaohao Wang, Kaiyang Ding, Kai Ni and Qian Zhou

Remote Sens. 2023, 15(23), 5607; https://doi.org/10.3390/rs15235607 - 2 Dec 2023

Cited by 1 | Viewed by 1316

Abstract

As marine activities expand, deploying underwater autonomous vehicles (AUVs) becomes critical. Efficiently navigating these AUVs through intricate underwater terrains is vital. This paper proposes a sophisticated motion-planning algorithm integrating deep reinforcement learning (DRL) with an improved artificial potential field (IAPF). The algorithm incorporates [...] Read more.

As marine activities expand, deploying underwater autonomous vehicles (AUVs) becomes critical. Efficiently navigating these AUVs through intricate underwater terrains is vital. This paper proposes a sophisticated motion-planning algorithm integrating deep reinforcement learning (DRL) with an improved artificial potential field (IAPF). The algorithm incorporates remote sensing information to overcome traditional APF challenges and combines the IAPF with the traveling salesman problem for optimal path cruising. Through a combination of DRL and multi-source data optimization, the approach ensures minimal energy consumption across all target points. Inertial sensors further refine trajectory, ensuring smooth navigation and precise positioning. The comparative experiments confirm the method’s energy efficiency, trajectory refinement, and safety excellence. Full article

(This article belongs to the Topic Artificial Intelligence in Navigation)

►▼ Show Figures

Figure 1

Figure 1
<p>Schematic of multi-source-data-assisted AUV for multiple target point cruising.</p> Full article ">Figure 2
<p>Kinematic and dynamic models for the AUV.</p> Full article ">Figure 3
<p>The improvement strategies for the local optimum problem. (<b>a</b>) The improvement strategy for a single obstacle in the 2D environment. (<b>b</b>) The improvement strategy for multiple obstacles in the 2D environment. (<b>c</b>) The improvement strategy for a single obstacle in the 3D environment. (<b>d</b>) The improvement strategy for multiple obstacles in the 3D environment.</p> Full article ">Figure 4
<p>Kalman filter process.</p> Full article ">Figure 5
<p>Multi-source-data-assisted AUV motion planning based on the DDPG algorithm.</p> Full article ">Figure 6
<p>Cruise sequence generated by TSP-IAPF.</p> Full article ">Figure 7
<p>Reward curves generated by four DRL algorithms.</p> Full article ">Figure 8
<p>The planned 3D paths from the four algorithms: (<b>a</b>) top view; (<b>b</b>) 3D view.</p> Full article ">Figure 9
<p>The four algorithms’ generated motion data. (<b>a</b>) Steering angle. (<b>b</b>) Attack angle. (<b>c</b>) Pitch angle. (<b>d</b>) Nearest distance to the obstacle. (<b>e</b>) Velocity. (<b>f</b>) Acceleration.</p> Full article ">Figure 10
<p>Energy consumption, the closest distance to the obstacle, path length, and navigation time for the four algorithms under Monte Carlo simulation. (<b>a</b>) Simulation of energy consumption and closest distance to the obstacle. (<b>b</b>) Simulation of path length and navigation time.</p> Full article ">Figure 11
<p>(<b>a</b>,<b>b</b>) Motion-planning paths are generated by four algorithms in a dynamic environment.</p> Full article ">Figure 12
<p>Trajectory-tracking curve for AUV under remote sensing error. (<b>a</b>) Top view of trajectory tracking curve. (<b>b</b>) Three-dimensional view of trajectory-tracking curve.</p> Full article ">Figure 13
<p>Energy consumption and closest distance to the obstacle for different remote sensing detection distances: (<b>a</b>) energy consumption; (<b>b</b>) nearest distance to the obstacle.</p> Full article ">Figure 14
<p>Trajectory-tracking error for AUV.</p> Full article ">

16 pages, 4228 KiB  

Open AccessArticle

Path Planning for Automatic Berthing Using Ship-Maneuvering Simulation-Based Deep Reinforcement Learning

by Anh Khoa Vo, Thi Loan Mai and Hyeon Kyu Yoon

Appl. Sci. 2023, 13(23), 12731; https://doi.org/10.3390/app132312731 - 27 Nov 2023

Cited by 3 | Viewed by 1440

Abstract

Despite receiving much attention from researchers in the field of naval architecture and marine engineering since the early stages of modern shipbuilding, the berthing phase is still one of the biggest challenges in ship maneuvering due to the potential risks involved. Many algorithms [...] Read more.

Despite receiving much attention from researchers in the field of naval architecture and marine engineering since the early stages of modern shipbuilding, the berthing phase is still one of the biggest challenges in ship maneuvering due to the potential risks involved. Many algorithms have been proposed to solve this problem. This paper proposes a new approach with a path-planning algorithm for automatic berthing tasks using deep reinforcement learning (RL) based on a maneuvering simulation. Unlike the conventional path-planning algorithm using the control theory or an advanced algorithm using deep learning, a state-of-the-art path-planning algorithm based on reinforcement learning automatically learns, explores, and optimizes the path for berthing performance through trial and error. The results of performing the twin delayed deep deterministic policy gradient (TD3) combined with the maneuvering simulation show that the approach can be used to propose a feasible and safe path for high-performing automatic berthing tasks. Full article

(This article belongs to the Topic Artificial Intelligence in Navigation)

►▼ Show Figures

Figure 1

Figure 1
<p>Coordinate system of the twin-propeller and twin-rudder ship model.</p> Full article ">Figure 2
<p>Geometry of an autonomous surface ship (KASS).</p> Full article ">Figure 3
<p>Simulation results of turning trajectories at a rudder angle of 35 degrees at three and six knots [<a href="#B20-applsci-13-12731" class="html-bibr">20</a>].</p> Full article ">Figure 4
<p>Simulation results of turning trajectories at a rudder angle of 35 degrees.</p> Full article ">Figure 5
<p>Satellite image of the Busan port.</p> Full article ">Figure 6
<p>Concept of the path-planning algorithm.</p> Full article ">Figure 7
<p>Reward coefficients.</p> Full article ">Figure 8
<p>Geometry of the port and berthing situation: (<b>a</b>) parallel berthing task (case 1) and (<b>b</b>) perpendicular berthing task (case 2).</p> Full article ">Figure 9
<p>Simulation results in a parallel berthing task (case 1).</p> Full article ">Figure 10
<p>Simulation results in the perpendicular berthing task (case 2).</p> Full article ">Figure 11
<p>Learning performance of TD3: (<b>a</b>) parallel berthing (case 1) and (<b>b</b>) perpendicular berthing task (case 2).</p> Full article ">

19 pages, 3947 KiB  

Open AccessTechnical Note

A Semantics-Guided Visual Simultaneous Localization and Mapping with U-Net for Complex Dynamic Indoor Environments

by Zhi Zeng, Hui Lin, Zhizhong Kang, Xiaokui Xie, Juntao Yang, Chuyu Li and Longze Zhu

Remote Sens. 2023, 15(23), 5479; https://doi.org/10.3390/rs15235479 - 23 Nov 2023

Cited by 2 | Viewed by 1195

Abstract

Traditional simultaneous localization and mapping (SLAM) system tends to operate in small-area static environments, and its performance might degrade when moving objects appear in a highly dynamic environment. To address this issue, this paper proposes a dynamic object-aware visual SLAM algorithm specifically designed [...] Read more.

Traditional simultaneous localization and mapping (SLAM) system tends to operate in small-area static environments, and its performance might degrade when moving objects appear in a highly dynamic environment. To address this issue, this paper proposes a dynamic object-aware visual SLAM algorithm specifically designed for dynamic indoor environments. The proposed method leverages a semantic segmentation architecture called U-Net, which is utilized in the tracking thread to detect potentially moving targets. The resulting output of semantic segmentation is tightly coupled with the geometric information extracted from the corresponding SLAM system, thus associating the feature points captured by images with the potentially moving targets. Finally, filtering out the moving feature points can greatly enhance localization accuracy in dynamic indoor environments. Quantitative and qualitative experiments were carried out on both the Technical University of Munich (TUM) public dataset and the real scenario dataset to verify the effectiveness and robustness of the proposed method. Results demonstrate that the semantics-guided approach significantly outperforms the ORB SLAM2 framework in dynamic indoor environments, which is crucial for improving the robustness and reliability of the SLAM system. Full article

(This article belongs to the Topic Artificial Intelligence in Navigation)

►▼ Show Figures

Figure 1

Figure 1
<p>Pipeline of the proposed method built on an ORB SLAM2 system. The three threads filled with gray are the same as the original threads of ORB SLAM2. The improvements over the original ORB SLAM2 system are highlighted in orange.</p> Full article ">Figure 2
<p>Framework of the U-Net architecture.</p> Full article ">Figure 3
<p>Camera trajectory in the first experiment of the real scenario. The red dotted line is the camera location in left or right to desk.</p> Full article ">Figure 4
<p>Camera trajectory in the second and third experiments of the real scenario. (<b>a</b>) Second experiment. (<b>b</b>) Third experiment.</p> Full article ">Figure 5
<p>Part of the VOC dataset.</p> Full article ">Figure 6
<p>Part of real scenario datasets.</p> Full article ">Figure 7
<p>Semantic segmentation results of the VOC. (<b>a</b>) Image #1, (<b>b</b>) ground truth #1, (<b>c</b>) predicted output #1, (<b>d</b>) image #2, (<b>e</b>) ground truth #2, (<b>f</b>) predicted output #2, (<b>g</b>) image #3, (<b>h</b>) ground truth #3, (<b>i</b>) predicted output #3, (<b>j</b>) image #4, (<b>k</b>) ground truth #4, (<b>l</b>) predicted output #4, (<b>o</b>) image #5, (<b>p</b>) ground truth #5, (<b>q</b>) predicted output #5, (<b>r</b>) image #6, (<b>s</b>) ground truth #6, (<b>t</b>) predicted output #6.</p> Full article ">Figure 8
<p>Semantic segmentation results of both TUM and real scenario datasets. (<b>a</b>) Image #1, (<b>b</b>) predicted output #1, (<b>c</b>) image #2, (<b>d</b>) predicted output #2, (<b>e</b>) image #3, (<b>f</b>) predicted output #3, (<b>g</b>) image #4, (<b>h</b>) predicted output #4, (<b>i</b>) image #3, (<b>j</b>) predicted output #3, (<b>k</b>) image #4, (<b>l</b>) predicted output #4.</p> Full article ">Figure 9
<p>The trajectory graphics on TUM datasets for different scenes. (<b>a</b>) Trajectory of fr3_sitting_static in a three-dimensional space. (<b>b</b>) Trajectory of fr3_walking_rpy in the XY direction. (<b>c</b>) Trajectory of fr3_walking_half in the XY direction.</p> Full article ">Figure 10
<p>Predicted trajectory of the teaching building datasets compared with ground truth. (<b>a</b>) First scenario, (<b>b</b>) second scenario, (<b>c</b>) third scenario.</p> Full article ">

27 pages, 6787 KiB  

Open AccessArticle

A Hybrid POI Recommendation System Combining Link Analysis and Collaborative Filtering Based on Various Visiting Behaviors

by Sumet Darapisut, Komate Amphawan, Nutthanon Leelathakul and Sunisa Rimcharoen

ISPRS Int. J. Geo-Inf. 2023, 12(10), 431; https://doi.org/10.3390/ijgi12100431 - 22 Oct 2023

Cited by 1 | Viewed by 1863

Abstract

Location-based recommender systems (LBRSs) have exhibited significant potential in providing personalized recommendations based on the user’s geographic location and contextual factors such as time, personal preference, and location categories. However, several challenges (such as data sparsity, the cold-start problem, and tedium problem) need [...] Read more.

Location-based recommender systems (LBRSs) have exhibited significant potential in providing personalized recommendations based on the user’s geographic location and contextual factors such as time, personal preference, and location categories. However, several challenges (such as data sparsity, the cold-start problem, and tedium problem) need to be addressed to develop more effective LBRSs. In this paper, we propose a novel POI recommendation system, called LACF-Rec3, which employs a hybrid approach of link analysis (HITS-3) and collaborative filtering (CF-3) based on three visiting behaviors: frequency, variety, and repetition. HITS-3 identifies distinctive POIs based on user- and POI-visit patterns, ranks them accordingly, and recommends them to cold-start users. For existing users, CF-3 utilizes collaborative filtering based on their previous check-in history and POI distinctive aspects. Our experimental results conducted on a Foursquare dataset demonstrate that LACF-Rec3 outperforms prior methods in terms of recommendation accuracy, ranking precision, and matching ratio. In addition, LACF-Rec3 effectively solves the challenges of data sparsity, the cold-start issue, and tedium problems for cold-start and existing users. These findings highlight the potential of LACF-Rec3 as a promising solution to the challenges encountered by LBRS. Full article

(This article belongs to the Topic Artificial Intelligence in Navigation)

►▼ Show Figures

Figure 1

Figure 1
<p>Our LACF-Rec3 method consists of two phases: offline phase with HITS-3, and online phase with CF-3.</p> Full article ">Figure 2
<p>Example of the offline phase in the LACF-Rec3.</p> Full article ">Figure 3
<p>Example of the online phase in LACF-Rec3.</p> Full article ">Figure 4
<p>Recommendation effectiveness evaluation method using MBR.</p> Full article ">Figure 5
<p>Precision metric for cold-start users with respect to the recommendation numbers.</p> Full article ">Figure 6
<p>Recall metric for cold-start users with respect to the recommendation numbers.</p> Full article ">Figure 7
<p>NDCG metric for cold-start users with respect to the recommendation numbers.</p> Full article ">Figure 8
<p>Matching ratio metric for cold-start users with respect to the recommendation numbers.</p> Full article ">Figure 9
<p>Precision metric for existing users with respect to the recommendation numbers.</p> Full article ">Figure 10
<p>Recall metric for existing users with respect to the recommendation numbers.</p> Full article ">Figure 11
<p>NDCG metric for existing users with respect to the recommendation numbers.</p> Full article ">Figure 12
<p>Matching ratio metric for existing users with respect to the recommendation numbers.</p> Full article ">

17 pages, 10552 KiB  

Open AccessArticle

Scaling-Invariant Max-Filtering Enhancement Transformers for Efficient Visual Tracking

by Zhen Chen, Xingzhong Xiong, Fanqin Meng, Xianbing Xiao and Jun Liu

Electronics 2023, 12(18), 3905; https://doi.org/10.3390/electronics12183905 - 15 Sep 2023

Cited by 1 | Viewed by 1094

Abstract

Real-time tracking is one of the most challenging problems in computer vision. Most Transformer-based trackers usually require expensive computational and storage power, which leads to these robust trackers being unable to achieve satisfactory real-time performance in resource-constrained devices. In this work, we propose [...] Read more.

Real-time tracking is one of the most challenging problems in computer vision. Most Transformer-based trackers usually require expensive computational and storage power, which leads to these robust trackers being unable to achieve satisfactory real-time performance in resource-constrained devices. In this work, we propose a lightweight tracker, AnteaTrack. To localize the target more accurately, this paper presents a scaling-invariant max-filtering operator. It uses local max-pooling to filter the suspected target portion in overlapping sliding windows for enhancement while suppressing the background. For a more compact target bounding-box, this paper presents an upsampling module based on Pixel-Shuffle to increase the fine-grained expression of target features. In addition, AnteaTrack can run in real time at 47 frames per second (FPS) on a CPU. We tested AnteaTrack on five datasets, and a large number of experiments showed that AnteaTrack provides the most efficient solution compared to the same type of CPU real-time trackers. Full article

(This article belongs to the Topic Artificial Intelligence in Navigation)

►▼ Show Figures

Figure 1

Figure 1
<p>Comparison of the tracking results of our tracker with the baseline E.T.Track [<a href="#B17-electronics-12-03905" class="html-bibr">17</a>]. The mosaics (from left to right) shown in the second row are the classifier output feature maps of E.T.Track, E.T.Track with MET, and AnteaTrack, with their tracking results in the yellow, blue, and red boxes at the bottom. The ground-truth labeled anteater is indicated by the green box.</p> Full article ">Figure 2
<p>The scaling-invariant max-filtering enhancement Transformer proposed in this work, which we name MET. Here, AAP stands for adaptive average pooling and FFN represents a feedforward network.</p> Full article ">Figure 3
<p>Standard attention (<b>left</b>) and exemplar attention mechanisms (<b>right</b>).</p> Full article ">Figure 4
<p>How scaling-invariant max-filtering works. The sliding window with stride <math display="inline"><semantics> <msub> <mi>s</mi> <mi>k</mi> </msub> </semantics></math> and spatial size <math display="inline"><semantics> <mrow> <mi>M</mi> <mo>×</mo> <mi>M</mi> </mrow> </semantics></math> (<span class="html-italic">M</span> is constrained to be odd) has a built-in max-pooling operation. The output samples the region of the left feature for the max-keep operator and then compares the features before and after the window processing, and the positions where the response is unchanged are maintained. At the same time, the other parts are suppressed and set to 0 (the green portion of the right side).</p> Full article ">Figure 5
<p>Flowchart of the tracker proposed in this work. The red box represents the prediction result obtained by the combination of bounding box regression and target branch.</p> Full article ">Figure 6
<p>Performance comparison with our tracker using the OTB100 [<a href="#B22-electronics-12-03905" class="html-bibr">22</a>] dataset. (<b>a</b>) Normalized precision plot, (<b>b</b>) precision plot, (<b>c</b>) success plot. AnteaTrack is depicted by the solid red line.</p> Full article ">Figure 7
<p>Performance comparison with our tracker using the UAV123 [<a href="#B23-electronics-12-03905" class="html-bibr">23</a>] dataset. (<b>a</b>) Normalized precision plot, (<b>b</b>) precision plot, (<b>c</b>) success plot. AnteaTrack is depicted by the solid red line.</p> Full article ">Figure 8
<p>Performance comparison with our tracker using the LaSOT [<a href="#B21-electronics-12-03905" class="html-bibr">21</a>] dataset. (<b>a</b>) Normalized precision plot, (<b>b</b>) precision plot, (<b>c</b>) success plot. AnteaTrack is depicted by the solid red line.</p> Full article ">Figure 9
<p>Performance comparison with our tracker using the NFS [<a href="#B24-electronics-12-03905" class="html-bibr">24</a>] dataset. (<b>a</b>) Normalized precision plot, (<b>b</b>) precision plot, (<b>c</b>) success plot. AnteaTrack is depicted by the solid red line.</p> Full article ">Figure 10
<p>Comparison of tracking results obtained by AnteaTrack and the baseline E.T.Track [<a href="#B17-electronics-12-03905" class="html-bibr">17</a>] with the NFS motion blur frames. The green box denotes the ground truth, the red box denotes AnteaTrack’s tracking results, and the yellow box represents E.T.Track’s.</p> Full article ">Figure 11
<p>Comparison of the effect of our tracker with the baseline using the LaSOT [<a href="#B21-electronics-12-03905" class="html-bibr">21</a>] and GOT-10K [<a href="#B25-electronics-12-03905" class="html-bibr">25</a>] partial sequences. The green box denotes the ground truth, and the red and yellow boxes denote the results of our tracker and E.T.Track [<a href="#B17-electronics-12-03905" class="html-bibr">17</a>], respectively. #1360 and basketball-7 denote frame 1360 of the basketball-7 sequence.</p> Full article ">

14 pages, 692 KiB  

Open AccessArticle

A Latent-Factor-Model-Based Approach for Traffic Data Imputation with Road Network Information

by Xing Su, Wenjie Sun, Chenting Song, Zhi Cai and Limin Guo

ISPRS Int. J. Geo-Inf. 2023, 12(9), 378; https://doi.org/10.3390/ijgi12090378 - 15 Sep 2023

Viewed by 1514

Abstract

With the rapid development of the economy, car ownership has grown rapidly, which causes many traffic problems. In recent years, intelligent transportation systems have been used to solve various traffic problems. To achieve effective and efficient traffic management, intelligent transportation systems need a [...] Read more.

With the rapid development of the economy, car ownership has grown rapidly, which causes many traffic problems. In recent years, intelligent transportation systems have been used to solve various traffic problems. To achieve effective and efficient traffic management, intelligent transportation systems need a large amount of complete traffic data. However, the current traffic data collection methods result in different forms of missing data. In the last twenty years, although many approaches have been proposed to impute missing data based on different mechanisms, these all have their limitations, which leads to low imputation accuracy, especially when the collected traffic data have a large amount of missing values. To this end, this paper proposes a latent-factor-model-based approach to impute the missing traffic data. In the proposed approach, the spatial information of the road network is first combined with the spatiotemporal matrix of the original traffic data. Then, the latent-factor-model-based algorithm is employed to impute the missing data in the combined matrix of the traffic data. Based on the real traffic data from METR-LA, we found that the imputation accuracy of the proposed approach was better than that of most of the current traffic-data-imputation approaches, especially when the original traffic data are limited. Full article

(This article belongs to the Topic Artificial Intelligence in Navigation)

►▼ Show Figures

Figure 1

Figure 1
<p>An example of a spatiotemporal matrix of original traffic data <math display="inline"><semantics> <mrow> <mi>V</mi> <mo>∈</mo> <msup> <mi mathvariant="double-struck">R</mi> <mrow> <mi>I</mi> <mo>·</mo> <mi>J</mi> </mrow> </msup> </mrow> </semantics></math>.</p> Full article ">Figure 2
<p>Three kinds of traffic data missing patterns.</p> Full article ">Figure 3
<p>The process of LFM based factorization of <span class="html-italic">X</span>.</p> Full article ">Figure 4
<p>An example of road network.</p> Full article ">Figure 5
<p>The original and normalized adjacent matrix of road network.</p> Full article ">Figure 6
<p>The combined matrix <math display="inline"><semantics> <mi mathvariant="italic">VD</mi> </semantics></math>.</p> Full article ">

17 pages, 8254 KiB  

Open AccessArticle

MCBM-SLAM: An Improved Mask-Region-Convolutional Neural Network-Based Simultaneous Localization and Mapping System for Dynamic Environments

by Xiankun Wang and Xinguang Zhang

Electronics 2023, 12(17), 3596; https://doi.org/10.3390/electronics12173596 - 25 Aug 2023

Cited by 2 | Viewed by 1234

Abstract

Current research on SLAM can be divided into two parts according to the research scenario: SLAM research in dynamic scenarios and SLAM research in static scenarios. Research is now relatively well established for static environments. However, in dynamic environments, the impact of moving [...] Read more.

Current research on SLAM can be divided into two parts according to the research scenario: SLAM research in dynamic scenarios and SLAM research in static scenarios. Research is now relatively well established for static environments. However, in dynamic environments, the impact of moving objects leads to inaccurate positioning accuracy and poor robustness of SLAM systems. To address the shortcomings of SLAM systems in dynamic environments, this paper develops a series of solutions to address these problems. First, an attention-based Mask R-CNN network is used to ensure the reliability of dynamic object extraction in dynamic environments. Dynamic feature points are then rejected based on the mask identified by the Mask R-CNN network, and a preliminary estimate of the camera pose is made. Secondly, in order to enhance the picture matching quality and efficiently reject the mismatched points, this paper proposes an image mismatching algorithm incorporating adaptive edge distance with grid motion statistics. Finally, static feature points on dynamic objects are re-added using motion constraints and chi-square tests, and the camera’s pose is re-estimated. The SLAM algorithm of this paper was run on the KITTI and TUM-RGBD datasets, respectively, and the results show that the SLAM algorithm of this paper outperforms the ORB-SLAM2 algorithm for sequences containing more dynamic objects in the KITTI dataset. On the TUM-RGBD dataset, the Dyna-SLAM algorithm increased localization accuracy by an average of 71.94% when compared to the ORB-SLAM2 method, while the SLAM algorithm in this study increased localization accuracy by an average of 78.18% when compared to the ORB-SLAM2 algorithm. When compared to the Dyna-SLAM technique, the SLAM algorithm in this work increased average positioning accuracy by 6.24%, proving that it is superior to Dyna-SLAM. Full article

(This article belongs to the Topic Artificial Intelligence in Navigation)

►▼ Show Figures

Figure 1

Figure 1
<p>Framework of SLAM system.</p> Full article ">Figure 2
<p>Processing of edge feature points.</p> Full article ">Figure 3
<p>Original GMS algorithm matching results.</p> Full article ">Figure 4
<p>Matching results of the improved GMS algorithm.</p> Full article ">Figure 5
<p>Basic structure of Mask R-CNN network.</p> Full article ">Figure 6
<p>CBAM-Mask network structure.</p> Full article ">Figure 7
<p>Spatial attention module.</p> Full article ">Figure 8
<p>Comparison of the results of adding attention and not adding attention (the <b>left column</b> has no added attention mechanism; the <b>right column</b> has added attention mechanism).</p> Full article ">Figure 9
<p>Motion consistency detection with cardinality experiment.</p> Full article ">Figure 10
<p>Results of feature point extraction after removing dynamic objects and without removing dynamic objects.</p> Full article ">Figure 11
<p>Comparison of the trajectories under the w_xyz sequence.</p> Full article ">Figure 12
<p>Comparison of the trajectories under the w_halfsphere sequence.</p> Full article ">Figure 13
<p>Comparison of the absolute trajectory errors of the three algorithms on the w_xyz sequence.</p> Full article ">Figure 14
<p>Comparison of the absolute trajectory errors of the three algorithms in the w_halfsphere sequence.</p> Full article ">Figure 15
<p>Comparison of trajectories on the KITTI07 sequence. (<b>a</b>) Comparison chart with real trajectory. (<b>b</b>) Comparison results in xyz direction. (<b>c</b>) Comparison results in the rpy direction.</p> Full article ">Figure 16
<p>Comparison of absolute trajectory error between ORB-SLAM2 and MCBM-SLAM algorithm.</p> Full article ">

16 pages, 1899 KiB  

Open AccessArticle

SASTGCN: A Self-Adaptive Spatio-Temporal Graph Convolutional Network for Traffic Prediction

by Wei Li, Xi Zhan, Xin Liu, Lei Zhang, Yu Pan and Zhisong Pan

ISPRS Int. J. Geo-Inf. 2023, 12(8), 346; https://doi.org/10.3390/ijgi12080346 - 18 Aug 2023

Cited by 1 | Viewed by 1544

Abstract

Traffic prediction plays a significant part in creating intelligent cities such as traffic management, urban computing, and public safety. Nevertheless, the complex spatio-temporal linkages and dynamically shifting patterns make it somewhat challenging. Existing mainstream traffic prediction approaches heavily rely on graph convolutional networks [...] Read more.

Traffic prediction plays a significant part in creating intelligent cities such as traffic management, urban computing, and public safety. Nevertheless, the complex spatio-temporal linkages and dynamically shifting patterns make it somewhat challenging. Existing mainstream traffic prediction approaches heavily rely on graph convolutional networks and sequence prediction methods to extract complicated spatio-temporal patterns statically. However, they neglect to account for dynamic underlying correlations and thus fail to produce satisfactory prediction results. Therefore, we propose a novel Self-Adaptive Spatio-Temporal Graph Convolutional Network (SASTGCN) for traffic prediction. A self-adaptive calibrator, a spatio-temporal feature extractor, and a predictor comprise the bulk of the framework. To extract the distribution bias of the input in the self-adaptive calibrator, we employ a self-supervisor made of an encoder–decoder structure. The concatenation of the bias and the original characteristics are provided as input to the spatio-temporal feature extractor, which leverages a transformer and graph convolution structures to learn the spatio-temporal pattern, and then applies a predictor to produce the final prediction. Extensive trials on two public traffic prediction datasets (METR-LA and PEMS-BAY) demonstrate that SASTGCN surpasses the most recent techniques in several metrics. Full article

(This article belongs to the Topic Artificial Intelligence in Navigation)

►▼ Show Figures

Figure 1

Figure 1
<p>Some applications of traffic prediction.</p> Full article ">Figure 2
<p>Graph-structured traffic data prediction.</p> Full article ">Figure 3
<p>The overall architecture of our proposed SASTGCN.</p> Full article ">Figure 4
<p>The influence of input horizon length <math display="inline"><semantics> <mi>H</mi> </semantics></math> on the METR-LA dataset.</p> Full article ">Figure A1
<p>The distribution of the selected sensors in the METR-LA dataset.</p> Full article ">

19 pages, 4200 KiB  

Open AccessArticle

An Improved Adaptive Sparrow Search Algorithm for TDOA-Based Localization

by Jiaqi Dong, Zengzeng Lian, Jingcheng Xu and Zhe Yue

ISPRS Int. J. Geo-Inf. 2023, 12(8), 334; https://doi.org/10.3390/ijgi12080334 - 9 Aug 2023

Cited by 1 | Viewed by 1497

Abstract

The Ultra-Wideband (UWB) indoor positioning method is widely used in areas where no satellite signals are available. However, during the measurement process of UWB, the collected data contain random errors. To alleviate the effect of random errors on positioning accuracy, an improved adaptive [...] Read more.

The Ultra-Wideband (UWB) indoor positioning method is widely used in areas where no satellite signals are available. However, during the measurement process of UWB, the collected data contain random errors. To alleviate the effect of random errors on positioning accuracy, an improved adaptive sparrow search algorithm (IASSA) based on the sparrow search algorithm (SSA) is proposed in this paper by introducing three strategies, namely, the two-step weighted least squares algorithm, adaptive adjustment of search boundary, and producer–scrounger quantity adaptive adjustment. The simulation and field test results indicate that the IASSA algorithm achieves significantly higher localization accuracy than previous methods. Meanwhile, the IASSA algorithm requires fewer iterations, which overcomes the problem of the long computation time of the swarm intelligence optimization algorithm. Therefore, the IASSA algorithm has advantages in indoor positioning accuracy and robustness performance. Full article

(This article belongs to the Topic Artificial Intelligence in Navigation)

►▼ Show Figures

Figure 1

Figure 1
<p>The TDOA positioning principle.</p> Full article ">Figure 2
<p>The difference between the result of the TSWLS algorithm and the real coordinates.</p> Full article ">Figure 3
<p>The distribution of the opposite of the fitness value.</p> Full article ">Figure 4
<p>The contour map and results of each algorithm solution.</p> Full article ">Figure 5
<p>Contour map and results of each algorithm solution.</p> Full article ">Figure 6
<p>The comparison of localization accuracy under different distance noise standard deviations.</p> Full article ">Figure 7
<p>The number of bad results for different algorithms.</p> Full article ">Figure 8
<p>The RMSE convergence of the GWO, SSA, and IASSA algorithms.</p> Full article ">Figure 9
<p>The experimental site and point distribution.</p> Full article ">Figure 10
<p>The algorithm errors at different points in the field experiment.</p> Full article ">Figure 11
<p>The trajectory tracking results of the five algorithms. (<b>a</b>) The trajectory tracking results of the TSWLS algorithm. (<b>b</b>) The trajectory tracking results of the ICWLS algorithm. (<b>c</b>) The trajectory tracking results of the GWO algorithm. (<b>d</b>) The trajectory tracking results of the SSA algorithm. (<b>e</b>) The trajectory tracking results of the IASSA algorithm.</p> Full article ">Figure 11 Cont.
<p>The trajectory tracking results of the five algorithms. (<b>a</b>) The trajectory tracking results of the TSWLS algorithm. (<b>b</b>) The trajectory tracking results of the ICWLS algorithm. (<b>c</b>) The trajectory tracking results of the GWO algorithm. (<b>d</b>) The trajectory tracking results of the SSA algorithm. (<b>e</b>) The trajectory tracking results of the IASSA algorithm.</p> Full article ">Figure 12
<p>CDF of position errors of different algorithms. (<b>a</b>) Before smoothing by the KF algorithm. (<b>b</b>) After smoothing by the KF algorithm.</p> Full article ">

14 pages, 6698 KiB  

Open AccessArticle

Skill Fusion in Hybrid Robotic Framework for Visual Object Goal Navigation

by Aleksei Staroverov, Kirill Muravyev, Konstantin Yakovlev and Aleksandr I. Panov

Robotics 2023, 12(4), 104; https://doi.org/10.3390/robotics12040104 - 16 Jul 2023

Cited by 2 | Viewed by 2465

Abstract

In recent years, Embodied AI has become one of the main topics in robotics. For the agent to operate in human-centric environments, it needs the ability to explore previously unseen areas and to navigate to objects that humans want the agent to interact [...] Read more.

In recent years, Embodied AI has become one of the main topics in robotics. For the agent to operate in human-centric environments, it needs the ability to explore previously unseen areas and to navigate to objects that humans want the agent to interact with. This task, which can be formulated as ObjectGoal Navigation (ObjectNav), is the main focus of this work. To solve this challenging problem, we suggest a hybrid framework consisting of both not-learnable and learnable modules and a switcher between them—SkillFusion. The former are more accurate, while the latter are more robust to sensors’ noise. To mitigate the sim-to-real gap, which often arises with learnable methods, we suggest training them in such a way that they are less environment-dependent. As a result, our method showed top results in both the Habitat simulator and during the evaluations on a real robot. Full article

(This article belongs to the Topic Artificial Intelligence in Navigation)

►▼ Show Figures

Figure 1

Figure 1
<p>A scheme of our object navigation pipeline used on a real robot. The pipeline consists of classical and RL-based parts, and each part has exploration and GoalReacher skills. To choose the appropriate skill at each moment, we implement the skill fusion decider, which selects an action from the skill with the highest score.</p> Full article ">Figure 2
<p>Examples of skill fusion work in a simulator. The red line is a trajectory that was attained by the classical pipeline, the light blue line was by RL exploration skill, and the dark blue was by RL GoalReacher skill.</p> Full article ">Figure 3
<p>An example of max pooling of the built map. Gray denotes an unknown area, black denotes obstacle cells, and white denotes free space cells.</p> Full article ">Figure 4
<p>The proposed neural net architecture for fusing navigation tasks at the RL training phase.</p> Full article ">Figure 5
<p>Example of skill management during the episode. The blue color of the trajectory is denoted RL Explore skill execution, dark blue denotes an RL GoalReacher skill, and a red color denotes classical skills.</p> Full article ">Figure 6
<p>Comparison of GoalReacher skill training using a frozen CLIP encoder versus an unfrozen ResNet encoder.</p> Full article ">Figure 7
<p>Robot platform based on the Clearpath Husky chassis with a ZED camera (<b>left</b>). We used it to evaluate our results in a real-world scene (<b>right</b>).</p> Full article ">Figure 8
<p>Robot trajectories: RL (<b>top</b>), classical pipeline (<b>middle</b>), and SkillFusion (<b>bottom</b>). The red box denotes the target object, the red circle denotes the start point, and the white circle with a blue arrow denotes the robot’s termination point and direction.</p> Full article ">

24 pages, 13149 KiB  

Open AccessArticle

An Improved Distributed Sampling PPO Algorithm Based on Beta Policy for Continuous Global Path Planning Scheme

by Qianhao Xiao, Li Jiang, Manman Wang and Xin Zhang

Sensors 2023, 23(13), 6101; https://doi.org/10.3390/s23136101 - 2 Jul 2023

Cited by 2 | Viewed by 2074

Abstract

Traditional path planning is mainly utilized for path planning in discrete action space, which results in incomplete ship navigation power propulsion strategies during the path search process. Moreover, reinforcement learning experiences low success rates due to its unbalanced sample collection and unreasonable design [...] Read more.

Traditional path planning is mainly utilized for path planning in discrete action space, which results in incomplete ship navigation power propulsion strategies during the path search process. Moreover, reinforcement learning experiences low success rates due to its unbalanced sample collection and unreasonable design of reward function. In this paper, an environment framework is designed, which is constructed using the Box2D physics engine and employs a reward function, with the distance between the agent and arrival point as the main, and the potential field superimposed by boundary control, obstacles, and arrival point as the supplement. We also employ the state-of-the-art PPO (Proximal Policy Optimization) algorithm as a baseline for global path planning to address the issue of incomplete ship navigation power propulsion strategy. Additionally, a Beta policy-based distributed sample collection PPO algorithm is proposed to overcome the problem of unbalanced sample collection in path planning by dividing sub-regions to achieve distributed sample collection. The experimental results show the following: (1) The distributed sample collection training policy exhibits stronger robustness in the PPO algorithm; (2) The introduced Beta policy for action sampling results in a higher path planning success rate and reward accumulation than the Gaussian policy at the same training time; (3) When planning a path of the same length, the proposed Beta policy-based distributed sample collection PPO algorithm generates a smoother path than traditional path planning algorithms, such as A*, IDA*, and Dijkstra. Full article

(This article belongs to the Topic Artificial Intelligence in Navigation)

►▼ Show Figures

Figure 1

Figure 1
<p>Location of simulation environment generation.</p> Full article ">Figure 2
<p>Data sequence state-space.</p> Full article ">Figure 3
<p>Potential-based reward: (<b>a</b>) Obstacle-based potential plane; (<b>b</b>) Arrival point-based potential plane; (<b>c</b>) Boundary control-based potential plane.</p> Full article ">Figure 4
<p>Potential plane based on obstacle, arrival point, and boundary control (superimposed).</p> Full article ">Figure 5
<p>Actor network structure.</p> Full article ">Figure 6
<p>Testing results: (<b>a</b>) PPO baseline planning test; (<b>b</b>) Distributed fixed starting position test.</p> Full article ">Figure 7
<p>Refine sub-regions.</p> Full article ">Figure 8
<p>Distributed PPO algorithm.</p> Full article ">Figure 9
<p>Gaussian boundary effects.</p> Full article ">Figure 10
<p>Accumulated rewards of PPO baseline compared with distributed sample extraction PPO.</p> Full article ">Figure 11
<p>Planning path comparison: (<b>a</b>) A* and RL search path in scenario1; (<b>b</b>) A* and RL search path in scenario2.</p> Full article ">Figure 12
<p>Sub-region accumulated rewards: (<b>a</b>) Data sequence state-space; (<b>b</b>) Image state-space.</p> Full article ">Figure 13
<p>Comparison of planning path length: (<b>a</b>) Data sequence state-space; (<b>b</b>) Image state-space.</p> Full article ">

21 pages, 5035 KiB  

Open AccessArticle

A New Geomagnetic Vector Navigation Method Based on a Two-Stage Neural Network

by Zhuo Chen, Zhongyan Liu, Qi Zhang, Dixiang Chen, Mengchun Pan and Yujing Xu

Electronics 2023, 12(9), 1975; https://doi.org/10.3390/electronics12091975 - 24 Apr 2023

Cited by 3 | Viewed by 1403

Abstract

The traditional geomagnetic matching navigation method is based on the correlation criteria operations between measurement sequences and a geomagnetic map. However, when the gradient of the geomagnetic field is small, there are multiple similar data in the geomagnetic database to the measurement value, [...] Read more.

The traditional geomagnetic matching navigation method is based on the correlation criteria operations between measurement sequences and a geomagnetic map. However, when the gradient of the geomagnetic field is small, there are multiple similar data in the geomagnetic database to the measurement value, which means the correlation-based matching method fails. Based on the idea of pattern recognition, this paper constructs a two-stage neural network by cascading a probabilistic neural network and a non-fully connected neural network to, respectively, classify geomagnetic vectors and their feature information in two steps: “coarse screening” and “fine screening”. The effectiveness and accuracy of the geomagnetic vector navigation algorithm based on the two-stage neural network are verified through simulation and experiments. In simulation, it is verified that when the geomagnetic average gradient is 5 nT/km, the traditional geomagnetic matching method fails, while the positioning accuracy based on the proposed method is 40.17 m, and the matching success rate also reaches 98.13%. Further, in flight experiments, under an average gradient of 11 nT/km, the positioning error based on the proposed method is 39.01 m, and the matching success rate also reaches 99.42%. Full article

(This article belongs to the Topic Artificial Intelligence in Navigation)

►▼ Show Figures

Figure 1

Figure 1
<p>Principles of geomagnetic pattern recognition navigation.</p> Full article ">Figure 2
<p>PNN model.</p> Full article ">Figure 3
<p>Positioning result of pattern recognition PNN.</p> Full article ">Figure 4
<p>Probabilistic top 10 patterns of pattern recognition navigation based on PNN.</p> Full article ">Figure 5
<p>Two-stage pattern recognition navigation algorithm flowchart.</p> Full article ">Figure 6
<p>Two-stage neural network model.</p> Full article ">Figure 7
<p>Sigmoid function.</p> Full article ">Figure 8
<p>ReLU function.</p> Full article ">Figure 9
<p>Sample set production principle.</p> Full article ">Figure 10
<p>Simulated geomagnetic field map.</p> Full article ">Figure 11
<p>Comparison of pattern recognition PNN and two-stage NN tracks.</p> Full article ">Figure 12
<p>Experimental aircraft with strapdown navigation system.</p> Full article ">Figure 13
<p>Geomagnetic field in the experiment region.</p> Full article ">Figure 14
<p>Comparison of four geomagnetic navigation methods based on flight survey data.</p> Full article ">

15 pages, 27174 KiB  

Open AccessArticle

Image Retrieval Method Based on Visual Map Pre-Sampling Construction in Indoor Positioning

by Jianan Bai, Danyang Qin, Ping Zheng and Lin Ma

ISPRS Int. J. Geo-Inf. 2023, 12(4), 169; https://doi.org/10.3390/ijgi12040169 - 14 Apr 2023

Viewed by 1563

Abstract

In visual indoor positioning systems, the method of constructing a visual map by point-by-point sampling is widely used due to its characteristics of clear static images and simple coordinate calculation. However, too small a sampling interval will cause image redundancy, while too large [...] Read more.

In visual indoor positioning systems, the method of constructing a visual map by point-by-point sampling is widely used due to its characteristics of clear static images and simple coordinate calculation. However, too small a sampling interval will cause image redundancy, while too large a sampling interval will lead to the absence of any scene images, which will result in worse positioning efficiency and inferior positioning accuracy. As a result, this paper proposed a visual map construction method based on pre-sampled image features matching, according to the epipolar geometry of adjacent position images, to determine the optimal sampling spacing within the constraints and effectively control the database size while ensuring the integrity of the image information. In addition, in order to realize the rapid retrieval of the visual map and reduce the positioning error caused by the time overhead, an image retrieval method based on deep hashing was also designed in this paper. This method used a convolutional neural network to extract image features to construct the semantic similarity structure to guide the generation of hash code. Based on the log-cosh function, this paper proposed a loss function whose function curve was smooth and not affected by outliers, and then integrated it into the deep network to optimize parameters, for fast and accurate image retrieval. Experiments on the FLICKR25K dataset and the visual map proved that the method proposed in this paper could achieve sub-second image retrieval with guaranteed accuracy, thereby demonstrating its promising performance. Full article

(This article belongs to the Topic Artificial Intelligence in Navigation)

►▼ Show Figures

Figure 1

Figure 1
<p>Roadmap of the research.</p> Full article ">Figure 2
<p>(<b>a</b>) DJI Pocket2 and laser rangefinder; (<b>b</b>) MATLAB Mobile.</p> Full article ">Figure 3
<p>(<b>a</b>) Image capture angle; (<b>b</b>) determination of image acquisition position.</p> Full article ">Figure 4
<p>Epipolar geometry.</p> Full article ">Figure 5
<p>Distribution of sampling points in the electronics laboratory building.</p> Full article ">Figure 6
<p>Flowchart of image retrieval based on deep hashing.</p> Full article ">Figure 7
<p>Histogram distribution of image cosine distances in the visual map.</p> Full article ">Figure 8
<p>mAP of three optimizers.</p> Full article ">Figure 9
<p>Example of the visual map.</p> Full article ">Figure 10
<p>Precision–recall curves and top-N precision with code length 32.</p> Full article ">Figure 11
<p>mAP on Visual Map.</p> Full article ">Figure 12
<p>Precision–recall curves and top-N precision with code length 32 on visual map.</p> Full article ">

Show export options expand_more Show export options expand_less

Select all

Export citation of selected articles as:

Plain Text BibTeX BibTeX (without abstracts) Endnote Endnote (without abstracts) Tab-delimited RIS

 

Error

Oops... you haven't selected anything for export.

Ok

clear

Displaying articles 1-18