[go: up one dir, main page]
More Web Proxy on the site http://driver.im/
You seem to have javascript disabled. Please note that many of the page functionalities won't work as expected without javascript enabled.
 
 
Sign in to use this feature.

Years

Between: -

Subjects

remove_circle_outline
remove_circle_outline
remove_circle_outline
remove_circle_outline
remove_circle_outline
remove_circle_outline
remove_circle_outline
remove_circle_outline
remove_circle_outline

Journals

remove_circle_outline
remove_circle_outline
remove_circle_outline

Article Types

Countries / Regions

remove_circle_outline
remove_circle_outline
remove_circle_outline
remove_circle_outline

Search Results (277)

Search Parameters:
Keywords = real-time multi-agents

Order results
Result details
Results per page
Select all
Export citation of selected articles as:
33 pages, 7735 KiB  
Article
Control and Optimization of Hydrogen Hybrid Electric Vehicles Using GPS-Based Speed Estimation
by Nouha Mansouri, Aymen Mnassri, Sihem Nasri, Majid Ali, Abderezak Lashab, Juan C. Vasquez and Josep M. Guerrero
Electronics 2025, 14(1), 110; https://doi.org/10.3390/electronics14010110 - 30 Dec 2024
Viewed by 525
Abstract
This paper investigates the feasibility of hydrogen-powered hybrid electric vehicles as a solution to transportation-related pollution. It focuses on optimizing energy use to improve efficiency and reduce emissions. The study details the creation and real-time performance assessment of a hydrogen hybrid electric vehicle [...] Read more.
This paper investigates the feasibility of hydrogen-powered hybrid electric vehicles as a solution to transportation-related pollution. It focuses on optimizing energy use to improve efficiency and reduce emissions. The study details the creation and real-time performance assessment of a hydrogen hybrid electric vehicle (HHEV)system using an STM32F407VG board. This system includes a fuel cell (FC) as the main energy source, a battery (Bat) to provide energy during hydrogen supply disruptions and a supercapacitor (SC) to handle power fluctuations. A multi-agent-based artificial intelligence tool is used to model the system components, and an energy management algorithm (EMA) is applied to optimize energy use and support decision-making. Real Global Positioning System (GPS) data are analyzed to estimate energy consumption based on trip and speed parameters. The EMA, developed and implemented in real-time using Matlab/Simulink(2016), identifies the most energy-efficient routes. The results show that the proposed vehicle architecture and management strategy effectively select optimal routes with minimal energy use. Full article
Show Figures

Figure 1

Figure 1
<p>Hydrogen hybrid electric vehicle design.</p>
Full article ">Figure 2
<p>Speed estimator.</p>
Full article ">Figure 3
<p>Road recognition by GPS.</p>
Full article ">Figure 4
<p>System control behavior based on decision-making.</p>
Full article ">Figure 5
<p>EMA state diagram.</p>
Full article ">Figure 6
<p>FC/BaT/SC configuration EMA.</p>
Full article ">Figure 7
<p>Role of supercapacitors in hybrid hydrogen–electric vehicles.</p>
Full article ">Figure 8
<p>Component sizing scheme.</p>
Full article ">Figure 9
<p>Scheme of system efficiency.</p>
Full article ">Figure 10
<p>Possible detected ways by Google Maps.</p>
Full article ">Figure 11
<p>Referential vehicle speed value per road.</p>
Full article ">Figure 11 Cont.
<p>Referential vehicle speed value per road.</p>
Full article ">Figure 12
<p>Comparison of actual and predicted speed profiles with statistical evaluation across roads.</p>
Full article ">Figure 13
<p>Road 1 performance simulation.</p>
Full article ">Figure 14
<p>Road 2 performance simulation.</p>
Full article ">Figure 15
<p>Road 3 performance simulation.</p>
Full article ">Figure 15 Cont.
<p>Road 3 performance simulation.</p>
Full article ">Figure 16
<p>HEV performances per road.</p>
Full article ">Figure 17
<p>Interactive real-time MATLAB interface.</p>
Full article ">
24 pages, 1649 KiB  
Article
Heterogeneous Multi-Agent Risk-Aware Graph Encoder with Continuous Parameterized Decoder for Autonomous Driving Trajectory Prediction
by Shaoyu Sun, Chunyang Wang, Bo Xiao, Xuelian Liu, Chunhao Shi, Rongliang Sun and Ruijie Han
Electronics 2025, 14(1), 105; https://doi.org/10.3390/electronics14010105 - 30 Dec 2024
Viewed by 254
Abstract
Trajectory prediction is a critical component of autonomous driving, intelligent transportation systems, and human–robot interactions, particularly in complex environments like intersections, where diverse road constraints and multi-agent interactions significantly increase the risk of collisions. To address these challenges, a Heterogeneous Risk-Aware Graph Encoder [...] Read more.
Trajectory prediction is a critical component of autonomous driving, intelligent transportation systems, and human–robot interactions, particularly in complex environments like intersections, where diverse road constraints and multi-agent interactions significantly increase the risk of collisions. To address these challenges, a Heterogeneous Risk-Aware Graph Encoder with Continuous Parameterized Decoder for Trajectory Prediction (HRGC) is proposed. The architecture integrates a heterogeneous risk-aware local graph attention encoder, a low-rank temporal transformer, a fusion lane and global interaction encoder layer, and a continuous parameterized decoder. First, a heterogeneous risk-aware edge-enhanced local attention encoder is proposed, which enhances edge features using risk metrics, constructs graph structures through graph optimization and spectral clustering, maps these enhanced edge features to corresponding graph structure indices, and enriches node features with local agent-to-agent attention. Risk-aware edge attention is aggregated to update node features, capturing spatial and collision-aware representations, embedding crucial risk information into agents’ features. Next, the low-rank temporal transformer is employed to reduce computational complexity while preserving accuracy. By modeling agent-to-lane relationships, it captures critical map context, enhancing the understanding of agent behavior. Global interaction further refines node-to-node interactions via attention mechanisms, integrating risk and spatial information for improved trajectory encoding. Finally, a trajectory decoder utilizes the aforementioned encoder to generate control points for continuous parameterized curves. These control points are multiplied by dynamically adjusted basis functions, which are determined by an adaptive knot vector that adjusts based on velocity and curvature. This mechanism ensures precise local control and the superior handling of sharp turns and speed variations, resulting in more accurate real-time predictions in complex scenarios. The HRGC network achieves superior performance on the Argoverse 1 benchmark, outperforming state-of-the-art methods in complex urban intersections. Full article
(This article belongs to the Section Artificial Intelligence)
Show Figures

Figure 1

Figure 1
<p>Autonomous driving trajectory prediction.</p>
Full article ">Figure 2
<p>The overall architecture of HRGC trajectory predictor. It comprises four main blocks, processing heterogeneous multi-agent local feature embedding through a risk-aware edge enhanced node graph attention. Initially, we process node representation with rotation matrix, then construct heterogeneous graph by designing and aggregating risk aware edge and update node with attention. Subsequently, we apply low-rank temporal transformer layer to extract temporal features. These features are then fed into agent to lane layer, which fuses agent and lane feature for better local scene feature embedding, and last global interaction layer to extract agent and lane local features. Finally, we utilize a continuous parameterized trajectory decoder to decode rich and accurate features, generating continuous parameterized trajectories.</p>
Full article ">Figure 3
<p>Risk-aware edge layer.</p>
Full article ">Figure 4
<p>Moving direction with velocity risk.</p>
Full article ">Figure 5
<p>Time to collision.</p>
Full article ">Figure 6
<p>Edge index via graph optimization and clustering. First, we use Gaussian kernel to compute every two-node similarity and obtain the adjacency matrix. We compute Laplacian matrix by Equation (<a href="#FD6-electronics-14-00105" class="html-disp-formula">6</a>); susbsequently, we use the minimum cut graph optimization operate on Laplacian matrix to obtain the cluster index of node <math display="inline"><semantics> <msubsup> <mi>C</mi> <mi>i</mi> <mi>t</mi> </msubsup> </semantics></math> and <math display="inline"><semantics> <msubsup> <mi>C</mi> <mi>j</mi> <mi>t</mi> </msubsup> </semantics></math>.</p>
Full article ">Figure 7
<p>B-spline with control points.</p>
Full article ">Figure 8
<p>Ablation study of decoder variants with continuous trajectory decoder.</p>
Full article ">Figure 9
<p>Inference speed and paramters with minADE comparison with state-of-the-art methods.</p>
Full article ">Figure 10
<p>The red column represents the intersection successful case, while the green column represents the continuous parameterized trajectory prediction performance analysis.</p>
Full article ">Figure 11
<p>Failure case.</p>
Full article ">
27 pages, 9470 KiB  
Article
Multi-Objective Dynamic Path Planning with Multi-Agent Deep Reinforcement Learning
by Mengxue Tao, Qiang Li and Junxi Yu
J. Mar. Sci. Eng. 2025, 13(1), 20; https://doi.org/10.3390/jmse13010020 - 27 Dec 2024
Viewed by 287
Abstract
Multi-agent reinforcement learning (MARL) is characterized by its simple structure and strong adaptability, which has led to its widespread application in the field of path planning. To address the challenge of optimal path planning for mobile agent clusters in uncertain environments, a multi-objective [...] Read more.
Multi-agent reinforcement learning (MARL) is characterized by its simple structure and strong adaptability, which has led to its widespread application in the field of path planning. To address the challenge of optimal path planning for mobile agent clusters in uncertain environments, a multi-objective dynamic path planning model (MODPP) based on multi-agent deep reinforcement learning (MADRL) has been proposed. This model is suitable for complex, unstable task environments characterized by dimensionality explosion and offers scalability. The approach consists of two components: an action evaluation module and an action decision module, utilizing a centralized training with decentralized execution (CTDE) training architecture. During the training process, agents within the cluster learn cooperative strategies while being able to communicate with one another. Consequently, they can navigate through task environments without communication, achieving collision-free paths that optimize multiple sub-objectives globally, minimizing time, distance, and overall costs associated with turning. Furthermore, in real-task execution, agents acting as mobile entities can perform real-time obstacle avoidance. Finally, based on the OpenAI Gym platform, environments such as simple multi-objective environment and complex multi-objective environment were designed to analyze the rationality and effectiveness of the multi-objective dynamic path planning through minimum cost and collision risk assessments. Additionally, the impact of reward function configuration on agent strategies was discussed. Full article
Show Figures

Figure 1

Figure 1
<p>The framework of MADRL based MODPP method.</p>
Full article ">Figure 2
<p>Multi-agent reinforcement learning task.</p>
Full article ">Figure 3
<p>State transitions of the environment.</p>
Full article ">Figure 4
<p>Action space.</p>
Full article ">Figure 5
<p>Safe distance of proximity penalty.</p>
Full article ">Figure 6
<p>The CTED framework.</p>
Full article ">Figure 7
<p>Training and decision planning.</p>
Full article ">Figure 8
<p>The (<b>a</b>,<b>c</b>,<b>e</b>,<b>g</b>) shows the action paths of the three agents, and the (<b>b</b>,<b>d</b>,<b>f</b>,<b>h</b>) shows a simulation of the agents’ actions. The stars represent the goal sites.</p>
Full article ">Figure 8 Cont.
<p>The (<b>a</b>,<b>c</b>,<b>e</b>,<b>g</b>) shows the action paths of the three agents, and the (<b>b</b>,<b>d</b>,<b>f</b>,<b>h</b>) shows a simulation of the agents’ actions. The stars represent the goal sites.</p>
Full article ">Figure 9
<p>The (<b>a</b>,<b>b</b>) shows the speed change in the three agents in the <span class="html-italic">x</span> and <span class="html-italic">y</span> direction.</p>
Full article ">Figure 10
<p>The rewards given by the environment to the agent change over time.</p>
Full article ">Figure 11
<p>The (<b>a</b>,<b>c</b>,<b>e</b>,<b>g</b>) shows the action paths of the three agents, and the (<b>b</b>,<b>d</b>,<b>f</b>,<b>h</b>) shows a simulation of the agents’ actions. The stars represent the goal sites.</p>
Full article ">Figure 11 Cont.
<p>The (<b>a</b>,<b>c</b>,<b>e</b>,<b>g</b>) shows the action paths of the three agents, and the (<b>b</b>,<b>d</b>,<b>f</b>,<b>h</b>) shows a simulation of the agents’ actions. The stars represent the goal sites.</p>
Full article ">Figure 12
<p>The (<b>a</b>,<b>b</b>) shows the speed change in the three agents in the <span class="html-italic">x</span> and <span class="html-italic">y</span> direction.</p>
Full article ">Figure 13
<p>The rewards given by the environment to the agent change over time.</p>
Full article ">Figure 14
<p>The Figures (<b>a</b>,<b>c</b>,<b>e</b>,<b>g</b>) show the action paths of the three agents, and the Figures (<b>b</b>,<b>d</b>,<b>f</b>,<b>h</b>) show simulations of the agents’ actions. The stars represent the goal sites.</p>
Full article ">Figure 14 Cont.
<p>The Figures (<b>a</b>,<b>c</b>,<b>e</b>,<b>g</b>) show the action paths of the three agents, and the Figures (<b>b</b>,<b>d</b>,<b>f</b>,<b>h</b>) show simulations of the agents’ actions. The stars represent the goal sites.</p>
Full article ">Figure 15
<p>The (<b>a</b>,<b>b</b>) shows the speed change in the three agents in the <span class="html-italic">x</span> and <span class="html-italic">y</span> direction.</p>
Full article ">Figure 16
<p>The reward given to the agent by the environment changes over time steps.</p>
Full article ">Figure 17
<p>(<b>a</b>) Effect comparison of four algorithms in a simple multi-objective environment; (<b>b</b>–<b>d</b>) The environmental reward received by the agents changes over the model training episodes.</p>
Full article ">
26 pages, 10534 KiB  
Article
Assessment of the Impact of Multi-Agent Model-Based Traffic Optimization Interventions on Urban Travel Behavior
by Lihu Pan, Nan Yang, Linliang Zhang, Rui Zhang, Binhong Xie and Huimin Yan
Electronics 2025, 14(1), 13; https://doi.org/10.3390/electronics14010013 - 24 Dec 2024
Viewed by 243
Abstract
With the continuous increase in car ownership, alleviating traffic congestion and reducing carbon emissions have become key challenges in urban traffic management. This study constructs a multi-agent model to evaluate the impact of various traffic optimization interventions on citizens’ travel behavior and traffic [...] Read more.
With the continuous increase in car ownership, alleviating traffic congestion and reducing carbon emissions have become key challenges in urban traffic management. This study constructs a multi-agent model to evaluate the impact of various traffic optimization interventions on citizens’ travel behavior and traffic carbon emission levels. Different from previous mathematical models, this model integrates computer technology and geographic information systems, abstracting travelers as agents with self-control capabilities who can make independent decisions based on their own circumstances, thus reflecting individual differences in travel behavior. Using the real geographical and social environment of the high-density travel area in Xiaodian District, Taiyuan City as a case study, this research explores the overall improvement in the urban transportation system through the implementation of multiple traffic optimization interventions, such as a parking reservation system, the promotion of the park-and-ride mode, and the optimization of public transportation services. Studies have demonstrated that, compared to reducing bus fares, travelers exhibit a greater sensitivity to waiting times. Reducing bus departure intervals can increase the proportion of park-and-ride trips to 25.79%, surpassing the 19.19% increase observed with fare adjustments. A moderate increase in the proportion of reserved parking spaces can elevate the public transport load to 49.85%. The synergistic effect of a combined strategy can further boost the public transport share to 50.62%, while increasing the park-and-ride trip proportion to 33.6%, thereby highlighting the comprehensive benefits of implementing multiple strategies in tandem. When the parking reservation system is effectively implemented, carbon dioxide emissions can be reduced from over 800 kg to below 200 kg, and the proportion of vehicle cruising can decrease from over 20% to under 15%. These results underscore the critical role of the parking reservation strategy in optimizing traffic flow and advancing environmental sustainability. Full article
Show Figures

Figure 1

Figure 1
<p>Target study area.</p>
Full article ">Figure 2
<p>Model framework.</p>
Full article ">Figure 3
<p>Model hierarchy.</p>
Full article ">Figure 4
<p>Real urban road network.</p>
Full article ">Figure 5
<p>System operation process.</p>
Full article ">Figure 6
<p>The impact of bus fares and frequency of departure on travel choices.</p>
Full article ">Figure 7
<p>The impact of bus frequency on average arrival time.</p>
Full article ">Figure 8
<p>The impact of the proportion of parking spaces that can be reserved on the uptake of park-and-ride and public transport.</p>
Full article ">Figure 9
<p>The impact of combination strategies on the choice of park-and-ride travel.</p>
Full article ">Figure 10
<p>The impact of a combination strategy on the average arrival time.</p>
Full article ">Figure 11
<p>The impact of combined strategies on public transport load.</p>
Full article ">Figure 12
<p>The impact of parking reservations on cruising time and cruising ratio.</p>
Full article ">Figure 13
<p>The impact of parking reservations on cruising distance and cruising speed.</p>
Full article ">Figure 14
<p>The impact of parking reservation on the average distance traveled by car and the public transport load factor.</p>
Full article ">Figure 15
<p>Average vehicle speed and average cruising time in reserved and non-reserved scenarios.</p>
Full article ">Figure 16
<p>Cruising ratio and cruising distance in reserved and non-reserved scenarios.</p>
Full article ">Figure 17
<p>Simulation of multiple processes in parallel.</p>
Full article ">
16 pages, 2276 KiB  
Article
Adaptive Control of VSG Inertia Damping Based on MADDPG
by Demu Zhang, Jing Zhang, Yu He, Tao Shen and Xingyan Liu
Energies 2024, 17(24), 6421; https://doi.org/10.3390/en17246421 - 20 Dec 2024
Viewed by 277
Abstract
As renewable energy sources become more integrated into the power grid, traditional virtual synchronous generator (VSG) control strategies have become inadequate for the current low-damping, low-inertia power systems. Therefore, this paper proposes a VSG inertia and damping adaptive control method based on multi-agent [...] Read more.
As renewable energy sources become more integrated into the power grid, traditional virtual synchronous generator (VSG) control strategies have become inadequate for the current low-damping, low-inertia power systems. Therefore, this paper proposes a VSG inertia and damping adaptive control method based on multi-agent deep deterministic policy gradient (MADDPG). The paper first introduces the working principles of virtual synchronous generators and establishes a corresponding VSG model. Based on this model, the influence of variations in virtual inertia (J) and damping (D) coefficients on fluctuations in active power output is examined, defining the action space for J and D. The proposed method is mainly divided into two phases: “centralized training and decentralized execution”. In the centralized training phase, each agent’s critic network shares global observation and action information to guide the actor network in policy optimization. In the decentralized execution phase, agents observe frequency deviations and the rate at which angular frequency changes, using reinforcement learning algorithms to adjust the virtual inertia J and damping coefficient D in real time. Finally, the effectiveness of the proposed MADDPG control strategy is validated through comparison with adaptive control and DDPG control methods. Full article
(This article belongs to the Special Issue Planning, Operation, and Control of New Power Systems)
Show Figures

Figure 1

Figure 1
<p>Topology diagram of the VSG principle.</p>
Full article ">Figure 2
<p>Equivalent circuit model.</p>
Full article ">Figure 3
<p>Active power response curve (<span class="html-italic">J</span> varies).</p>
Full article ">Figure 4
<p>Active power response curve (<span class="html-italic">D</span> varies).</p>
Full article ">Figure 5
<p>Active and reactive power control loop.</p>
Full article ">Figure 6
<p>MADDPG parameters are self-adapting.</p>
Full article ">Figure 7
<p>Pre-training active power fluctuation chart.</p>
Full article ">Figure 8
<p>Training results chart.</p>
Full article ">Figure 9
<p>Case 1 frequency response curve chart.</p>
Full article ">Figure 10
<p>Case 1 power response curve chart.</p>
Full article ">Figure 11
<p>Case 2 frequency response curve chart.</p>
Full article ">Figure 12
<p><span class="html-italic">J</span> and <span class="html-italic">D</span> variation chart.</p>
Full article ">
25 pages, 6743 KiB  
Article
Online Autonomous Motion Control of Communication-Relay UAV with Channel Prediction in Dynamic Urban Environments
by Cancan Tao and Bowen Liu
Drones 2024, 8(12), 771; https://doi.org/10.3390/drones8120771 - 19 Dec 2024
Viewed by 474
Abstract
In order to improve the network performance of multi-unmanned ground vehicle (UGV) systems in urban environments, this article proposes a novel online autonomous motion-control method for the relay UAV. The problem is solved by jointly considering unknown RF channel parameters, unknown multi-agent mobility, [...] Read more.
In order to improve the network performance of multi-unmanned ground vehicle (UGV) systems in urban environments, this article proposes a novel online autonomous motion-control method for the relay UAV. The problem is solved by jointly considering unknown RF channel parameters, unknown multi-agent mobility, the impact of the environments on channel characteristics, and the unavailable angle-of-arrival (AoA) information of the received signal, making the solution of the problem more practical and comprehensive. The method mainly consists of two parts: wireless channel parameter estimation and optimal relay position search. Considering that in practical applications, the radio frequency (RF) channel parameters in complex urban environments are difficult to obtain in advance and are constantly changing, an estimation algorithm based on Gaussian process learning is proposed for online evaluation of the wireless channel parameters near the current position of the UAV; for the optimal relay position search problem, in order to improve the real-time performance of the method, a line search algorithm and a general gradient-based algorithm are proposed, which are used for point-to-point communication and multi-node communication scenarios, respectively, reducing the two-dimensional search to a one-dimensional search, and the stability proof and convergence conditions of the algorithm are given. Comparative experiments and simulation results under different scenarios show that the proposed motion-control method can drive the UAV to reach or track the optimal relay position and improve the network performance, while demonstrating that it is beneficial to consider the impact of the environments on the channel characteristics. Full article
Show Figures

Figure 1

Figure 1
<p>Illustration of air-to-ground relay communication scenario in urban environments.</p>
Full article ">Figure 2
<p>Motion control framework.</p>
Full article ">Figure 3
<p>Schematic diagram of air-to-ground signal propagation.</p>
Full article ">Figure 4
<p>Flight trajectories of the UAV that supports communication for two stationary UGVs.</p>
Full article ">Figure 5
<p>Changes in communication performance when the UAV supports communication for two stationary UGVs.</p>
Full article ">Figure 6
<p>Flight trajectories of the UAV that supports communication for multiple stationary UGVs.</p>
Full article ">Figure 7
<p>Changes in communication performance when the UAV supports communication for multiple stationary UGVs.</p>
Full article ">Figure 8
<p>Flight trajectories of the UAV that supports point-to-point communication for two moving UGVs.</p>
Full article ">Figure 9
<p>Changes in communication performance when the UAV supports point-to-point communication for two moving UGVs.</p>
Full article ">Figure 10
<p>Flight trajectories of the UAV that supports multi-node communication for multiple moving UGVs.</p>
Full article ">Figure 11
<p>Changes in communication performance when the UAV supports multi-node communication for multiple moving UGVs.</p>
Full article ">Figure 12
<p>Flight trajectories of the UAV that supports point-to-point communication for two moving UGVs with unknown channel parameters.</p>
Full article ">Figure 13
<p>Changes in communication performance when the UAV supports point-to-point communication for two moving UGVs with unknown channel parameters.</p>
Full article ">Figure 14
<p>Flight trajectories of the UAV that supports multi-node communication for multiple moving UGVs with unknown channel parameters.</p>
Full article ">Figure 15
<p>Changes in communication performance when the UAV supports multi-node communication for multiple moving UGVs with unknown channel parameters.</p>
Full article ">
17 pages, 2088 KiB  
Article
Personalized Clustering for Emotion Recognition Improvement
by Laura Gutiérrez-Martín, Celia López-Ongil, Jose M. Lanza-Gutiérrez and Jose A. Miranda Calero
Sensors 2024, 24(24), 8110; https://doi.org/10.3390/s24248110 - 19 Dec 2024
Viewed by 340
Abstract
Emotion recognition through artificial intelligence and smart sensing of physical and physiological signals (affective computing) is achieving very interesting results in terms of accuracy, inference times, and user-independent models. In this sense, there are applications related to the safety and well-being of people [...] Read more.
Emotion recognition through artificial intelligence and smart sensing of physical and physiological signals (affective computing) is achieving very interesting results in terms of accuracy, inference times, and user-independent models. In this sense, there are applications related to the safety and well-being of people (sexual assaults, gender-based violence, children and elderly abuse, mental health, etc.) that require even more improvements. Emotion detection should be done with fast, discrete, and non-luxurious systems working in real time and real life (wearable devices, wireless communications, battery-powered). Furthermore, emotional reactions to violence are not equal in all people. Then, large general models cannot be applied to a multi-user system for people protection, and health and social workers and law enforcement agents would welcome customized and lightweight AI models. These semi-personalized models will be applicable to clusters of subjects sharing similarities in their emotional reactions to external stimuli. This customization requires several steps: creating clusters of subjects with similar behaviors, creating AI models for every cluster, continually updating these models with new data, and enrolling new subjects in clusters when required. An initial approach for clustering labeled data compiled (physiological data, together with emotional labels) is presented in this work, as well as the method to ensure the enrollment of new users with unlabeled data once the AI models are generated. The idea is that this complete methodology can be exportable to any other expert systems where unlabeled data are added during in-field operation and different profiles exist in terms of data. Experimental results demonstrate an improvement of 5% in accuracy and 4% in F1 score with respect to our baseline general model, along with a 32% to 58% reduction in variability, respectively. Full article
Show Figures

Figure 1

Figure 1
<p>User profile clustering based on labeled observations training and testing scheme.</p>
Full article ">Figure 2
<p>Unlabeled observation clustering assignment scheme.</p>
Full article ">Figure 3
<p>Scheme of phase 1 and 2 of the experimental procedure for evaluating the impact of the methodologies on fear detection.</p>
Full article ">Figure 4
<p>Scheme of phase 3 of the experimental procedure for evaluating the impact of the methodologies on fear detection.</p>
Full article ">Figure 5
<p>Average performance metrics (accuracy and F1 score) per typology cluster with a parameter sweep for the general model contribution threshold. 0: only personalized model; 1: only general model.</p>
Full article ">
26 pages, 6416 KiB  
Article
Advanced Monocular Outdoor Pose Estimation in Autonomous Systems: Leveraging Optical Flow, Depth Estimation, and Semantic Segmentation with Dynamic Object Removal
by Alireza Ghasemieh and Rasha Kashef
Sensors 2024, 24(24), 8040; https://doi.org/10.3390/s24248040 - 17 Dec 2024
Viewed by 394
Abstract
Autonomous technologies have revolutionized transportation, military operations, and space exploration, necessitating precise localization in environments where traditional GPS-based systems are unreliable or unavailable. While widespread for outdoor localization, GPS systems face limitations in obstructed environments such as dense urban areas, forests, and indoor [...] Read more.
Autonomous technologies have revolutionized transportation, military operations, and space exploration, necessitating precise localization in environments where traditional GPS-based systems are unreliable or unavailable. While widespread for outdoor localization, GPS systems face limitations in obstructed environments such as dense urban areas, forests, and indoor spaces. Moreover, GPS reliance introduces vulnerabilities to signal disruptions, which can lead to significant operational failures. Hence, developing alternative localization techniques that do not depend on external signals is essential, showing a critical need for robust, GPS-independent localization solutions adaptable to different applications, ranging from Earth-based autonomous vehicles to robotic missions on Mars. This paper addresses these challenges using Visual odometry (VO) to estimate a camera’s pose by analyzing captured image sequences in GPS-denied areas tailored for autonomous vehicles (AVs), where safety and real-time decision-making are paramount. Extensive research has been dedicated to pose estimation using LiDAR or stereo cameras, which, despite their accuracy, are constrained by weight, cost, and complexity. In contrast, monocular vision is practical and cost-effective, making it a popular choice for drones, cars, and autonomous vehicles. However, robust and reliable monocular pose estimation models remain underexplored. This research aims to fill this gap by developing a novel adaptive framework for outdoor pose estimation and safe navigation using enhanced visual odometry systems with monocular cameras, especially for applications where deploying additional sensors is not feasible due to cost or physical constraints. This framework is designed to be adaptable across different vehicles and platforms, ensuring accurate and reliable pose estimation. We integrate advanced control theory to provide safety guarantees for motion control, ensuring that the AV can react safely to the imminent hazards and unknown trajectories of nearby traffic agents. The focus is on creating an AI-driven model(s) that meets the performance standards of multi-sensor systems while leveraging the inherent advantages of monocular vision. This research uses state-of-the-art machine learning techniques to advance visual odometry’s technical capabilities and ensure its adaptability across different platforms, cameras, and environments. By merging cutting-edge visual odometry techniques with robust control theory, our approach enhances both the safety and performance of AVs in complex traffic situations, directly addressing the challenge of safe and adaptive navigation. Experimental results on the KITTI odometry dataset demonstrate a significant improvement in pose estimation accuracy, offering a cost-effective and robust solution for real-world applications. Full article
(This article belongs to the Special Issue Sensors for Object Detection, Pose Estimation, and 3D Reconstruction)
Show Figures

Figure 1

Figure 1
<p>Proposed Pipeline Architecture.</p>
Full article ">Figure 2
<p>Optical flow processed output sample for one sequence of frames.</p>
Full article ">Figure 3
<p>Sample output of depth estimation.</p>
Full article ">Figure 4
<p>Sample output of the semantic segmentation.</p>
Full article ">Figure 5
<p>Sample output of dynamic object and sky removal.</p>
Full article ">Figure 6
<p>Step-by-step preprocessing samples.</p>
Full article ">Figure 7
<p>Pose estimator architecture. I changed it and replaced the image.</p>
Full article ">Figure 8
<p>Train/Loss chart for the KITTI odometry dataset.</p>
Full article ">Figure 9
<p>The validation/Loss chart for the KITTI odometry dataset shows that the pose estimator can learn more rapidly by providing extra scene information, especially semantic segmentation, to add correction weight to each class of objects and remove dynamic ones from the estimations.</p>
Full article ">Figure 10
<p>Train and validation loss for different preprocessing stages, including no preprocessing, OF, OF with depth estimation, and OF with depth and semantic segmentation.</p>
Full article ">Figure 11
<p>Proposed model’s tracking experience output for the KITTI odometry dataset. The <span class="html-italic">X</span> and <span class="html-italic">Y</span>-axis units are in meters.</p>
Full article ">Figure 12
<p>Train/Loss with different learning rates.</p>
Full article ">Figure 13
<p>Validation/Loss with different learning rates.</p>
Full article ">
26 pages, 3702 KiB  
Article
Real-Time Scheduling with Independent Evaluators: Explainable Multi-Agent Approach
by Artem Isakov, Danil Peregorodiev, Ivan Tomilov, Chuyang Ye, Natalia Gusarova, Aleksandra Vatian and Alexander Boukhanovsky
Technologies 2024, 12(12), 259; https://doi.org/10.3390/technologies12120259 - 14 Dec 2024
Viewed by 652
Abstract
This study introduces a multi-agent reinforcement learning approach to address the challenges of real-time scheduling in dynamic environments, with a specific focus on healthcare operations. The proposed system integrates the Human-in-the-Loop (HITL) paradigm, providing continuous feedback from human evaluators, and it employs a [...] Read more.
This study introduces a multi-agent reinforcement learning approach to address the challenges of real-time scheduling in dynamic environments, with a specific focus on healthcare operations. The proposed system integrates the Human-in-the-Loop (HITL) paradigm, providing continuous feedback from human evaluators, and it employs a sophisticated reward function to attenuate the effects of human-driven events. Novel mapping between reinforcement learning (RL) concepts and the Belief–Desire–Intention (BDI) framework is developed to enhance the explainability of the agent’s decision-making. A system is designed to adapt to changes in patient conditions and preferences while minimizing disruptions to existing schedules. Experimental results show a notable decrease in patient waiting times compared to conventional methods while adhering to operator-induced constraints. This approach offers a robust, explainable, and adaptable solution for the challenging tasks of scheduling in the environments that require human-centered decision-making. Full article
(This article belongs to the Section Information and Communication Technologies)
Show Figures

Figure 1

Figure 1
<p>State space: (<b>a</b>) special case with uniform target distribution across the planning horizon with fixed <math display="inline"><semantics> <mi>λ</mi> </semantics></math>. (<b>b</b>) General case with different target <math display="inline"><semantics> <mi>λ</mi> </semantics></math> levels across the planning horizon.</p>
Full article ">Figure 2
<p>Time shift in the planning horizon.</p>
Full article ">Figure 3
<p>Cyclic action space for a one-week-long planning horizon. Starting from Tuesday, the agent can move its bid to Monday, to Wednesday, or leave it as-is. The transition between Monday and Sunday provides an illustrative example of how prohibited actions are addressed. Remaining on Monday may result in a collision due to the lack of available space to shift towards the left boundary of the planning horizon. Consequently, we incorporated a transition from Monday to Sunday to address this issue. The same applies to the last day of the planning horizon.</p>
Full article ">Figure 4
<p>Observation space: (<b>a</b>) <span class="html-italic">w</span> is set to 3, which implies that the agent observes the environment state for three days in a row, and it can also see its absolute position in the planning horizon. (<b>b</b>) Some agents may need more information to learn the optimal policy; therefore, we deliberately broaden the agent window size, as shown here. The longer it takes to learn the optimal policy, the wider the required window size. This provides the agents with more information; however, it also results in higher penalties, as shown later.</p>
Full article ">Figure 5
<p>Patient’s health state.</p>
Full article ">Figure 6
<p>Reward function design: (<b>a</b>) the initial <math display="inline"><semantics> <mi>α</mi> </semantics></math> level for the Levy alpha-stable distribution is set to 2, which makes it close to a Gaussian distribution with location <math display="inline"><semantics> <mrow> <mi>μ</mi> <mo>=</mo> <mi>d</mi> </mrow> </semantics></math>, scale <math display="inline"><semantics> <mrow> <mi>σ</mi> <mo>=</mo> <mn>1</mn> </mrow> </semantics></math>, and skewness <math display="inline"><semantics> <mrow> <mi>β</mi> <mo>=</mo> <mn>0</mn> </mrow> </semantics></math>. (<b>b</b>) However, the operator may want to increase the window parameter <span class="html-italic">w</span> while decreasing the <math display="inline"><semantics> <mi>α</mi> </semantics></math> level. A slightly decreasing <math display="inline"><semantics> <mi>α</mi> </semantics></math> level will move the curve towards Cauchy distribution.</p>
Full article ">Figure 7
<p>Successive re-scheduling.</p>
Full article ">Figure 8
<p>Belief–desire–intention terminology mapping.</p>
Full article ">Figure 9
<p>Metric charts: (<b>a</b>) Error metrics of critic networks. (<b>b</b>) Actor networks. (<b>c</b>) Average episodic reward during model training.</p>
Full article ">Figure 10
<p>A four-step game episode.</p>
Full article ">Figure 11
<p>Patient allocation during iterative rescheduling: (<b>a</b>) 1st episode out of an 8-episode game (<b>b</b>) 4th episode out of an 8-episode game (<b>c</b>) 8th episode out of an 8-episode game.</p>
Full article ">Figure 12
<p>Prompt sensitivity analysis.</p>
Full article ">Figure 13
<p>Direct preference optimization.</p>
Full article ">Figure 14
<p>Day and time exploration space.</p>
Full article ">Figure A1
<p>Multi-level explainability in the human–agent systems.</p>
Full article ">Figure A2
<p>Chain of thoughts for interpreting system logs.</p>
Full article ">Figure A3
<p>Chain of thoughts used to analyze prompts sensitivity.</p>
Full article ">
17 pages, 6290 KiB  
Article
Real-Time Detection of IoT Anomalies and Intrusion Data in Smart Cities Using Multi-Agent System
by Maria Viorela Muntean
Sensors 2024, 24(24), 7886; https://doi.org/10.3390/s24247886 - 10 Dec 2024
Viewed by 461
Abstract
Analyzing IoT data is an important challenge in the smart cities domain due to the complexity of network traffic generated by a large number of interconnected devices: smart cameras, light bulbs, motion sensors, voice assistants, and so on. To overcome this issue, a [...] Read more.
Analyzing IoT data is an important challenge in the smart cities domain due to the complexity of network traffic generated by a large number of interconnected devices: smart cameras, light bulbs, motion sensors, voice assistants, and so on. To overcome this issue, a multi-agent system is proposed to deal with all machine learning steps, from preprocessing and labeling data to discovering the most suitable model for the analyzed dataset. This paper shows that dividing the work into different tasks, managed by specialized agents, and evaluating the discovered models by an Expert System Agent leads to better results in the learning process. Full article
(This article belongs to the Special Issue Advanced IoT Systems in Smart Cities: 2nd Edition)
Show Figures

Figure 1

Figure 1
<p>Multi-agent system architecture for IoT data.</p>
Full article ">Figure 2
<p>Data preprocessing system architecture for IoT data.</p>
Full article ">Figure 3
<p>Data classification system architecture for IoT data.</p>
Full article ">Figure 4
<p>The architecture for IoT data generation and collection proposed in the VARIoT project [<a href="#B21-sensors-24-07886" class="html-bibr">21</a>].</p>
Full article ">Figure 5
<p>Sample of initial data.</p>
Full article ">Figure 6
<p>Sample of preprocessed data.</p>
Full article ">Figure 7
<p>IoT Dataset distribution.</p>
Full article ">Figure 8
<p>Classification accuracy for different models.</p>
Full article ">Figure 9
<p>Time taken to build models.</p>
Full article ">Figure 10
<p>Classification accuracy for different k values.</p>
Full article ">Figure 11
<p>True negative rates for different k values.</p>
Full article ">Figure 12
<p>Time taken to build a model for different k values (seconds).</p>
Full article ">Figure 13
<p>Classification accuracy for different distance functions.</p>
Full article ">Figure 14
<p>True negative rates for different distance functions.</p>
Full article ">Figure 15
<p>Time taken to build models for different distance functions.</p>
Full article ">Figure 16
<p>Accuracy and TN rate for different cost matrices.</p>
Full article ">Figure 17
<p>Time taken to build models for different cost matrices.</p>
Full article ">
23 pages, 6025 KiB  
Article
Integrating Vision and Olfaction via Multi-Modal LLM for Robotic Odor Source Localization
by Sunzid Hassan, Lingxiao Wang and Khan Raqib Mahmud
Sensors 2024, 24(24), 7875; https://doi.org/10.3390/s24247875 - 10 Dec 2024
Viewed by 539
Abstract
Odor source localization (OSL) technology allows autonomous agents like mobile robots to localize a target odor source in an unknown environment. This is achieved by an OSL navigation algorithm that processes an agent’s sensor readings to calculate action commands to guide the robot [...] Read more.
Odor source localization (OSL) technology allows autonomous agents like mobile robots to localize a target odor source in an unknown environment. This is achieved by an OSL navigation algorithm that processes an agent’s sensor readings to calculate action commands to guide the robot to locate the odor source. Compared to traditional ‘olfaction-only’ OSL algorithms, our proposed OSL algorithm integrates vision and olfaction sensor modalities to localize odor sources even if olfaction sensing is disrupted by non-unidirectional airflow or vision sensing is impaired by environmental complexities. The algorithm leverages the zero-shot multi-modal reasoning capabilities of large language models (LLMs), negating the requirement of manual knowledge encoding or custom-trained supervised learning models. A key feature of the proposed algorithm is the ‘High-level Reasoning’ module, which encodes the olfaction and vision sensor data into a multi-modal prompt and instructs the LLM to employ a hierarchical reasoning process to select an appropriate high-level navigation behavior. Subsequently, the ‘Low-level Action’ module translates the selected high-level navigation behavior into low-level action commands that can be executed by the mobile robot. To validate our algorithm, we implemented it on a mobile robot in a real-world environment with non-unidirectional airflow environments and obstacles to mimic a complex, practical search environment. We compared the performance of our proposed algorithm to single-sensory-modality-based ‘olfaction-only’ and ‘vision-only’ navigation algorithms, and a supervised learning-based ‘vision and olfaction fusion’ (Fusion) navigation algorithm. The experimental results show that the proposed LLM-based algorithm outperformed the other algorithms in terms of success rates and average search times in both unidirectional and non-unidirectional airflow environments. Full article
Show Figures

Figure 1

Figure 1
<p>Flow diagram of the OSL system. The robot platform is equipped with a camera for vision and a chemical detector and an anemometer for olfactory sensing. The proposed algorithm utilizes a multi-modal LLM for navigation decision making.</p>
Full article ">Figure 2
<p>The framework of the proposed multi-modal LLM-based navigation algorithm. The three main modules are the ‘Environment Sensing’ module, ‘High-level Reasoning’ module, and ‘Low-level Action’ module.</p>
Full article ">Figure 3
<p>Robot notation. Robot position <math display="inline"><semantics> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> </semantics></math> and heading <math display="inline"><semantics> <mi>ψ</mi> </semantics></math> are monitored by the built-in localization system. Wind speed <span class="html-italic">u</span> and wind direction are measured from the additional anemometer in the body frame. Wind direction in the inertial frame <math display="inline"><semantics> <msub> <mi>ϕ</mi> <mrow> <mi>I</mi> <mi>n</mi> <mi>e</mi> <mi>r</mi> <mi>t</mi> <mi>i</mi> <mi>a</mi> <mi>l</mi> </mrow> </msub> </semantics></math> is derived from robot heading <math display="inline"><semantics> <mi>ψ</mi> </semantics></math> and wind direction in the body frame.</p>
Full article ">Figure 4
<p>Implementation of the prompt. The system prompt includes the task, actions, hints and output instructions. The final prompt (orange box) includes the system prompt (green box) and the olfactory description (blue box).</p>
Full article ">Figure 5
<p>Querying the LLM with image and prompt. The input of the model is the visual frame and the prompt. The output of the model is the high-level action selection.</p>
Full article ">Figure 6
<p>The flow diagram of the ‘High-level Reasoning’ module. It illustrates how the proposed LLM-based agent integrates visual and olfactory sensory observations to make high-level navigation behavior decisions.</p>
Full article ">Figure 7
<p>(<b>a</b>) Moth mate-seeking behaviors. This figure was retrieved from [<a href="#B73-sensors-24-07875" class="html-bibr">73</a>]. (<b>b</b>) Moth-inspired ‘surge’ and (<b>c</b>) ‘casting’ navigation behaviors.</p>
Full article ">Figure 8
<p>(<b>a</b>) Figure of the search area. The size of the search area is 8.2 m × 3.3 m. The odor source is a humidifier that generates ethanol plumes. An obstacle prevents vision of the plume initially and obstructs navigation. Two perpendicular electric fans are used to create unidirectional or non−unidirectional airflow. There are objects to test the visual reasoning capability of the LLM model. (<b>b</b>) Schematic diagram of the search area. We selected four different robot initial positions in the downwind area in the repeated tests.</p>
Full article ">Figure 9
<p>(<b>a</b>) The robot platform includes a camera for vision sensing and a chemical sensor and an anemometer for olfaction sensing. (<b>b</b>) The computation system consists of the robot platform and a remote PC. The dotted line represents a wireless link and the solid line represents a physical connection.</p>
Full article ">Figure 10
<p>Trajectory graph of a successful sample run with the proposed multi-modal LLM-based OSL algorithm in unidirectional airflow environment. The navigation behaviors are color-separated. The obstacle is indicated by an orange box, and the odor source is represented by a red point with the surrounding circular source declaration region.</p>
Full article ">Figure 11
<p>Examples of ‘environment sensing’ and ‘reasoning output’ by the GPT-4o model.</p>
Full article ">Figure 12
<p>Robot trajectories of repeated tests in unidirectional airflow environment: (<b>a</b>–<b>d</b>) ‘olfaction-only’ (OO); (<b>e</b>–<b>h</b>) ‘vision-only’ (VO); (<b>i</b>–<b>l</b>) ‘vision and olfaction fusion’ (Fusion); and (<b>m</b>–<b>p</b>) ‘LLM-based’ (LLM) navigation algorithms.</p>
Full article ">Figure 13
<p>Robot trajectories of repeated tests in non-unidirectional airflow environment: (<b>a</b>–<b>d</b>) ‘olfaction-only’ (OO); (<b>e</b>–<b>h</b>) ‘vision-only’ (VO); (<b>i</b>–<b>l</b>) ‘vision and olfaction fusion’ (Fusion); and (<b>m</b>–<b>p</b>) ‘LLM-based’ (LLM) navigation algorithms.</p>
Full article ">Figure 14
<p>Mean differences of success rates of the four navigation algorithms. The positive differences are statistically significant at family-wise error rate (FWER) of <math display="inline"><semantics> <mrow> <mn>5</mn> <mo>%</mo> </mrow> </semantics></math>.</p>
Full article ">
28 pages, 5225 KiB  
Article
MAARS: Multiagent Actor–Critic Approach for Resource Allocation and Network Slicing in Multiaccess Edge Computing
by Ducsun Lim and Inwhee Joe
Sensors 2024, 24(23), 7760; https://doi.org/10.3390/s24237760 - 4 Dec 2024
Viewed by 569
Abstract
This paper presents a novel algorithm to address resource allocation and network-slicing challenges in multiaccess edge computing (MEC) networks. Network slicing divides a physical network into virtual slices, each tailored to efficiently allocate resources and meet diverse service requirements. To maximize the completion [...] Read more.
This paper presents a novel algorithm to address resource allocation and network-slicing challenges in multiaccess edge computing (MEC) networks. Network slicing divides a physical network into virtual slices, each tailored to efficiently allocate resources and meet diverse service requirements. To maximize the completion rate of user-computing tasks within these slices, the problem is decomposed into two subproblems: efficient core-to-edge slicing (ECS) and autonomous resource slicing (ARS). ECS facilitates collaborative resource distribution through cooperation among edge servers, while ARS dynamically manages resources based on real-time network conditions. The proposed solution, a multiagent actor–critic resource scheduling (MAARS) algorithm, employs a reinforcement learning framework. Specifically, MAARS utilizes a multiagent deep deterministic policy gradient (MADDPG) for efficient resource distribution in ECS and a soft actor–critic (SAC) technique for robust real-time resource management in ARS. Simulation results demonstrate that MAARS outperforms benchmark algorithms, including heuristic-based, DQN-based, and A2C-based methods, in terms of task completion rates, resource utilization, and convergence speed. Thus, this study offers a scalable and efficient framework for resource optimization and network slicing in MEC networks, providing practical benefits for real-world deployments and setting a new performance benchmark in dynamic environments. Full article
(This article belongs to the Special Issue Sensing and Mobile Edge Computing)
Show Figures

Figure 1

Figure 1
<p>System architecture.</p>
Full article ">Figure 2
<p>Illustration of the MADDPG-based ECS algorithm.</p>
Full article ">Figure 3
<p>Task-completion ratio vs. arrival rate. RAM: resource-allocation management.</p>
Full article ">Figure 4
<p>Task-completion ratio vs. CPU frequency.</p>
Full article ">Figure 5
<p>Task-completion ratio vs. bandwidth.</p>
Full article ">Figure 6
<p>Reward values vs. number of iterations.</p>
Full article ">Figure 7
<p>Task-completion ratio vs. arrival rate.</p>
Full article ">Figure 8
<p>Task-completion ratio vs. CPU frequency.</p>
Full article ">Figure 9
<p>Task-completion ratio vs. bandwidth.</p>
Full article ">Figure 10
<p>Utility-function values vs. weight vectors.</p>
Full article ">Figure 11
<p>Loss ratio vs. number of iterations.</p>
Full article ">
17 pages, 2270 KiB  
Article
Fast Parameter Estimation of Linear Frequency Modulation Signals in Marine Environments Based on Gradient Optimization Strategy
by Jiawei Wen, Zhe Ouyang, Donghu Nie and Cong Ren
J. Mar. Sci. Eng. 2024, 12(12), 2195; https://doi.org/10.3390/jmse12122195 - 1 Dec 2024
Viewed by 485
Abstract
Multi-buoy sonar systems achieve target localization by receiving broadband Linear Frequency Modulation signals emitted from the transmitter. Accurate estimations of the parameters of Linear Frequency Modulation signals significantly enhance the localization accuracy. Linear Frequency Modulation signals can be focused into the fractional domain [...] Read more.
Multi-buoy sonar systems achieve target localization by receiving broadband Linear Frequency Modulation signals emitted from the transmitter. Accurate estimations of the parameters of Linear Frequency Modulation signals significantly enhance the localization accuracy. Linear Frequency Modulation signals can be focused into the fractional domain through Fractional Fourier Transform, but this increases the computational complexity. In marine environments, the low signal-to-noise ratio and multipath effects degrade the parameter estimation accuracy further. To address these issues, this paper proposes a fast estimation algorithm based on the Fractional Fourier Transform and a Gradient Subtraction-Average-Based Optimizer. This algorithm leverages the impulsive characteristics of Linear Frequency Modulation signals after Fractional Fourier Transform transformation, using the Fractional Fourier Transform as the fitness function. The Gradient Subtraction-Average-Based Optimizer algorithm includes three enhancement strategies: chaotic mapping initialization, a Golden Sine Algorithm, and an adaptive t-distribution variational operator. The simulation results demonstrate that the Gradient Subtraction-Average-Based Optimizer algorithm improves the issues of low diversity in the search agents, imbalanced global and local search capabilities, and susceptibility to local optima. A comparative analysis and statistical testing confirm that under a low signal-to-noise ratio and multipath effect conditions, the Gradient Subtraction-Average-Based Optimizer algorithm not only ensures real-time parameter estimation but also improves the estimation accuracy. The results of the parameter estimation provide reliable data support for subsequent target localization. Full article
(This article belongs to the Section Ocean Engineering)
Show Figures

Figure 1

Figure 1
<p>Schematic diagram of time-frequency plane rotation.</p>
Full article ">Figure 2
<p>Schematic of the FRFT time-frequency surface of a multipath LFM signal.</p>
Full article ">Figure 3
<p>Schematic of the “<math display="inline"><semantics> <mrow> <mo>⊝</mo> <mo>−</mo> </mrow> </semantics></math>subtraction” detection and development phase.</p>
Full article ">Figure 4
<p>GSABO pretreatment Flowchart.</p>
Full article ">Figure 5
<p>Flowchart of the optimization phase of the GSABO algorithm.</p>
Full article ">Figure 6
<p>Logistic−chaotic−sequence distribution with Lyapunov exponential map.</p>
Full article ">Figure 7
<p>Improved−Logistic−tent chaotic sequence distribution with Lyapunov exponential map.</p>
Full article ">Figure 8
<p>FRFT processes LFM signals.</p>
Full article ">Figure 9
<p>Average fitness convergence curves of different algorithms.</p>
Full article ">Figure 10
<p>Average−fitness convergence curves of different SNR algorithms.</p>
Full article ">Figure 11
<p>Average−relative error curves for different algorithms.</p>
Full article ">
17 pages, 2075 KiB  
Article
Extending Conflict-Based Search for Optimal and Efficient Quadrotor Swarm Motion Planning
by Zihao Wang, Zhiwei Zhang, Wenying Dou, Guangpeng Hu, Lifu Zhang and Meng Zhang
Drones 2024, 8(12), 719; https://doi.org/10.3390/drones8120719 - 29 Nov 2024
Viewed by 382
Abstract
Multi-agent pathfinding has been extensively studied by the robotics and artificial intelligence communities. The classical algorithm, conflict-based search (CBS), is widely used in various real-world applications due to its ability to solve large-scale conflict-free paths. However, classical CBS assumes discrete time–space planning and [...] Read more.
Multi-agent pathfinding has been extensively studied by the robotics and artificial intelligence communities. The classical algorithm, conflict-based search (CBS), is widely used in various real-world applications due to its ability to solve large-scale conflict-free paths. However, classical CBS assumes discrete time–space planning and overlooks physical constraints in actual scenarios, making it unsuitable for direct application in unmanned aerial vehicle (UAV) swarm. Inspired by the decentralized planning and centralized conflict resolution ideas of CBS, we propose, for the first time, an optimal and efficient UAV swarm motion planner that integrates state lattice with CBS without any underlying assumption, named SL-CBS. SL-CBS is a two-layer search algorithm: (1) The low-level search utilizes an improved state lattice. We design emergency stop motion primitives to ensure complete UAV dynamics and handle spatio-temporal constraints from high-level conflicts. (2) The high-level algorithm defines comprehensive conflict types and proposes a motion primitive conflict detection method with linear time complexity based on Sturm’s theory. Additionally, our modified independence detection (ID) technique is applied to enable parallel conflict processing. We validate the planning capabilities of SL-CBS in classical scenarios and compare these with the latest state-of-the-art (SOTA) algorithms, showing great improvements in success rate, computation time, and flight time. Finally, we conduct large-scale tests to analyze the performance boundaries of SL-CBS+ID. Full article
(This article belongs to the Section Drone Design and Development)
Show Figures

Figure 1

Figure 1
<p>Planning process of the SL-CBS algorithm. The black border represents the map boundary, the light green rectangles denote static obstacles, the blue hollow circles indicate the initial positions of the UAVs, and the cyan hollow circles represent the UAVs’ target regions. The magenta curves depict the planned UAV flight trajectories, with the solid red circles on the curves indicating the UAV states. (<b>a</b>) initial state; (<b>b</b>) the first iteration; and (<b>c</b>) the second iteration.</p>
Full article ">Figure 2
<p>Two types of conflicts. In (<b>a</b>), a state conflict occurs at the extended state at time <math display="inline"><semantics> <mrow> <mi>t</mi> <mo>+</mo> <mi>τ</mi> </mrow> </semantics></math>, while in (<b>b</b>), a motion primitive conflict occurs at some location within <math display="inline"><semantics> <mrow> <mo>(</mo> <mi>t</mi> <mo>,</mo> <mi>t</mi> <mo>+</mo> <mi>τ</mi> <mo>)</mo> </mrow> </semantics></math>. Here, <math display="inline"><semantics> <mi>τ</mi> </semantics></math> is time interval.</p>
Full article ">Figure 3
<p>Emergency stop motion primitive model.</p>
Full article ">Figure 4
<p>UAV swarm conflict graph. The magenta double-headed arrows represent the existence of conflicts between the trajectories of the UAVs.</p>
Full article ">Figure 5
<p>Flight trajectories of UAV swarm in classic scenario test instances. The black border represents the map boundary, while the light green areas indicate static obstacles. The magenta curves depict the flight trajectories of the UAVs. In (<b>a</b>), the green and orange curves also represent UAV flight trajectories. In (<b>b</b>–<b>d</b>), the solid red circles indicate the flight states of the UAV at intermediate times.</p>
Full article ">Figure 6
<p>Flight trajectories planned by K-CBS, db-CBS, and SL-CBS for swarm size <math display="inline"><semantics> <mrow> <mi>K</mi> <mo>=</mo> <mn>8</mn> </mrow> </semantics></math>. The black border represents the map boundaries, while the 20 gray circles indicate obstacle regions. The solid circles in various colors depict the current states of the UAV swarm, and the curves in different colors represent the trajectories already flown by the UAV swarm. (<b>a</b>) K-CBS; (<b>b</b>) db-CBS; and (<b>c</b>) SL-CBS.</p>
Full article ">Figure 7
<p>Flight trajectories planned by SL-CBS for a swarm size of <math display="inline"><semantics> <mrow> <mi>K</mi> <mo>=</mo> <mn>20</mn> </mrow> </semantics></math>. The black border represents the map boundary, while the 25 gray circles indicate obstacle regions. The solid circles in various colors represent the current states of the UAV swarm, with black arrows on the circles indicating the yaw direction of each UAV. The curves in different colors illustrate the trajectories already traveled by the UAV swarm. (<b>a</b>) <math display="inline"><semantics> <mrow> <mi>K</mi> <mo>=</mo> <mn>20</mn> </mrow> </semantics></math> (without yaw); and (<b>b</b>) <math display="inline"><semantics> <mrow> <mi>K</mi> <mo>=</mo> <mn>20</mn> </mrow> </semantics></math> (with yaw).</p>
Full article ">Figure 8
<p>Evaluation metrics results of SL-CBS under different swarm sizes: (<b>a</b>) success rate; (<b>b</b>) computation time; (<b>c</b>) total flight time; and (<b>d</b>) makespan.</p>
Full article ">
20 pages, 1304 KiB  
Article
Robust Reinforcement Learning Strategies with Evolving Curriculum for Efficient Bus Operations in Smart Cities
by Yuhan Tang, Ao Qu, Xuan Jiang, Baichuan Mo, Shangqing Cao, Joseph Rodriguez, Haris N Koutsopoulos, Cathy Wu and Jinhua Zhao
Smart Cities 2024, 7(6), 3658-3677; https://doi.org/10.3390/smartcities7060141 - 29 Nov 2024
Viewed by 711
Abstract
Public transit systems are critical to the quality of urban life, and enhancing their efficiency is essential for building cost-effective and sustainable smart cities. Historically, researchers sought reinforcement learning (RL) applications to mitigate bus bunching issues with holding strategies. Nonetheless, these attempts often [...] Read more.
Public transit systems are critical to the quality of urban life, and enhancing their efficiency is essential for building cost-effective and sustainable smart cities. Historically, researchers sought reinforcement learning (RL) applications to mitigate bus bunching issues with holding strategies. Nonetheless, these attempts often led to oversimplifications and misalignment with the goal of reducing the total time passengers spent in the system, resulting in less robust or non-optimal solutions. In this study, we introduce a novel setting where each bus, supervised by an RL agent, can appropriately form aggregated policies from three strategies (holding, skipping station, and turning around to serve the opposite direction). It’s difficult to learn them all together, due to learning complexity, we employ domain knowledge and develop a gradually expanding action space curriculum, enabling agents to learn these strategies incrementally. We incorporate Long Short-Term Memory (LSTM) in our model considering the temporal interrelation among these actions. To address the inherent uncertainties of real-world traffic systems, we impose Domain Randomization (DR) on variables such as passenger demand and bus schedules. We conduct extensive numerical experiments with the integration of synthetic and real-world data to evaluate our model. Our methodology proves effective, enhancing bus schedule reliability and reducing total passenger waiting time by over 15%, thereby improving bus operation efficiency and smoothering operations of buses that align with sustainable goals. This work highlights the potential of robust RL combined with curriculum learning for optimizing public transport in smart cities, offering a scalable solution for real-world multi-agent systems. Full article
(This article belongs to the Special Issue Cost-Effective Transportation Planning for Smart Cities)
Show Figures

Figure 1

Figure 1
<p>Delay Passing Down Caused by Holding.</p>
Full article ">Figure 2
<p>Representationof the Bus Line simulated in Simpy.</p>
Full article ">Figure 3
<p>The Training Environment.</p>
Full article ">Figure 4
<p>The Actor-Critic Framework for PPO Algorithm.</p>
Full article ">Figure 5
<p>Thereward plots of different scenarios with three strategies implemented.</p>
Full article ">Figure 6
<p>Bus Arrival Time Before and After Training.</p>
Full article ">Figure 7
<p>The evaluation of DR.</p>
Full article ">
Back to TopTop