Pub Date : 2024-02-20DOI: 10.1109/TIV.2024.3367654
Yilin Wang;Yiheng Feng
Model-based and learning-based methods are two main approaches modeling car-following behaviors. To combine advantages from both types of models, this study introduces a novel approach, IDM-Follower, which generates a sequence of the following vehicle's trajectory using a recurrent autoencoder informed by a physical car-following model, the Intelligent Driving Model (IDM). We design an innovative neural network (NN) structure with two independent encoders and an attention-based decoder to predict the trajectory sequence. The loss function accounts for discrepancies from both the physical car-following model and the NN predictions. Numerical experiments are conducted using simulated and real world (i.e., NGSIM) datasets under different data noise levels with varying weights between the learning loss and the model loss. Testing results show the proposed approach outperforms both model-based and learning-based baselines under real and high noise levels. The optimal integrating weight between the model and learning component is significantly influenced by data quality, which affects both prediction accuracy and safety metrics.
{"title":"IDM-Follower: A Model-Informed Deep Learning Method for Car-Following Trajectory Prediction","authors":"Yilin Wang;Yiheng Feng","doi":"10.1109/TIV.2024.3367654","DOIUrl":"https://doi.org/10.1109/TIV.2024.3367654","url":null,"abstract":"Model-based and learning-based methods are two main approaches modeling car-following behaviors. To combine advantages from both types of models, this study introduces a novel approach, IDM-Follower, which generates a sequence of the following vehicle's trajectory using a recurrent autoencoder informed by a physical car-following model, the Intelligent Driving Model (IDM). We design an innovative neural network (NN) structure with two independent encoders and an attention-based decoder to predict the trajectory sequence. The loss function accounts for discrepancies from both the physical car-following model and the NN predictions. Numerical experiments are conducted using simulated and real world (i.e., NGSIM) datasets under different data noise levels with varying weights between the learning loss and the model loss. Testing results show the proposed approach outperforms both model-based and learning-based baselines under real and high noise levels. The optimal integrating weight between the model and learning component is significantly influenced by data quality, which affects both prediction accuracy and safety metrics.","PeriodicalId":36532,"journal":{"name":"IEEE Transactions on Intelligent Vehicles","volume":"9 6","pages":"5014-5020"},"PeriodicalIF":14.0,"publicationDate":"2024-02-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141965413","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-02-20DOI: 10.1109/TIV.2024.3367919
Di Chen;Meixin Zhu;Hao Yang;Xuesong Wang;Yinhai Wang
Autonomous vehicles (AVs) have the potential to significantly revolutionize society by providing a secure and efficient mode of transportation. Recent years have witnessed notable advancements in autonomous driving perception and prediction, but the challenge of validating the performance of AVs remains largely unresolved. Data-driven microscopic traffic simulation has become an important tool for autonomous driving testing due to 1) availability of high-fidelity traffic data; 2) its advantages of enabling large-scale testing and scenario reproducibility; and 3) its potential in reactive and realistic traffic simulation. However, a comprehensive review of this topic is currently lacking. This paper aims to fill this gap by summarizing relevant studies. The primary objective of this paper is to review current research efforts and provide a futuristic perspective that will benefit future developments in the field. It introduces the general issues of data-driven traffic simulation and outlines key concepts and terms. After overviewing traffic simulation, various datasets and evaluation metrics commonly used are reviewed. The paper then offers a comprehensive evaluation of imitation learning, reinforcement learning, deep generative and deep learning methods, summarizing each and analyzing their advantages and disadvantages in detail. Moreover, it evaluates the state-of-the-art, existing challenges, and future research directions.
{"title":"Data-Driven Traffic Simulation: A Comprehensive Review","authors":"Di Chen;Meixin Zhu;Hao Yang;Xuesong Wang;Yinhai Wang","doi":"10.1109/TIV.2024.3367919","DOIUrl":"https://doi.org/10.1109/TIV.2024.3367919","url":null,"abstract":"Autonomous vehicles (AVs) have the potential to significantly revolutionize society by providing a secure and efficient mode of transportation. Recent years have witnessed notable advancements in autonomous driving perception and prediction, but the challenge of validating the performance of AVs remains largely unresolved. Data-driven microscopic traffic simulation has become an important tool for autonomous driving testing due to 1) availability of high-fidelity traffic data; 2) its advantages of enabling large-scale testing and scenario reproducibility; and 3) its potential in reactive and realistic traffic simulation. However, a comprehensive review of this topic is currently lacking. This paper aims to fill this gap by summarizing relevant studies. The primary objective of this paper is to review current research efforts and provide a futuristic perspective that will benefit future developments in the field. It introduces the general issues of data-driven traffic simulation and outlines key concepts and terms. After overviewing traffic simulation, various datasets and evaluation metrics commonly used are reviewed. The paper then offers a comprehensive evaluation of imitation learning, reinforcement learning, deep generative and deep learning methods, summarizing each and analyzing their advantages and disadvantages in detail. Moreover, it evaluates the state-of-the-art, existing challenges, and future research directions.","PeriodicalId":36532,"journal":{"name":"IEEE Transactions on Intelligent Vehicles","volume":"9 4","pages":"4730-4748"},"PeriodicalIF":8.2,"publicationDate":"2024-02-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141315237","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-02-19DOI: 10.1109/TIV.2024.3365997
Long Chen;Yuchen Li;Wushour Silamu;Qingquan Li;Shirong Ge;Fei-Yue Wang
The increasing importance of mineral resources in contemporary society is becoming more prominent, playing an indispensable and crucial role in the global economy. These resources not only provide essential raw materials for the global economic system but also play an irreplaceable role in supporting the development of modern industry, technology, and infrastructure. With the rapid development of intelligent technologies such as Industry 5.0 and advanced Large Language Models (LLMs), the mining industry is facing unprecedented opportunities and challenges. The development of smart mines has become a crucial direction for industry progress. This article aims to explore the strategic requirements for the development of smart mines by combining advanced products or technologies such as Chat-GPT (one of the successful applications of LLMs), digital twins, and scenario engineering. We propose a comprehensive architecture consisting of three different levels, the mining industrial Internet of Things (IoT) platform, mining operating systems, and foundation models. The systems and models empower the mining equipment for transportation. The architecture delivers a comprehensive solution that aligns perfectly with the demands of Industry 5.0. The application and validation outcomes of this intelligent solution showcase a noteworthy enhancement in mining efficiency and a reduction in safety risks, thereby laying a sturdy groundwork for the advent of Mining 5.0.
{"title":"Smart Mining With Autonomous Driving in Industry 5.0: Architectures, Platforms, Operating Systems, Foundation Models, and Applications","authors":"Long Chen;Yuchen Li;Wushour Silamu;Qingquan Li;Shirong Ge;Fei-Yue Wang","doi":"10.1109/TIV.2024.3365997","DOIUrl":"https://doi.org/10.1109/TIV.2024.3365997","url":null,"abstract":"The increasing importance of mineral resources in contemporary society is becoming more prominent, playing an indispensable and crucial role in the global economy. These resources not only provide essential raw materials for the global economic system but also play an irreplaceable role in supporting the development of modern industry, technology, and infrastructure. With the rapid development of intelligent technologies such as Industry 5.0 and advanced Large Language Models (LLMs), the mining industry is facing unprecedented opportunities and challenges. The development of smart mines has become a crucial direction for industry progress. This article aims to explore the strategic requirements for the development of smart mines by combining advanced products or technologies such as Chat-GPT (one of the successful applications of LLMs), digital twins, and scenario engineering. We propose a comprehensive architecture consisting of three different levels, the mining industrial Internet of Things (IoT) platform, mining operating systems, and foundation models. The systems and models empower the mining equipment for transportation. The architecture delivers a comprehensive solution that aligns perfectly with the demands of Industry 5.0. The application and validation outcomes of this intelligent solution showcase a noteworthy enhancement in mining efficiency and a reduction in safety risks, thereby laying a sturdy groundwork for the advent of Mining 5.0.","PeriodicalId":36532,"journal":{"name":"IEEE Transactions on Intelligent Vehicles","volume":"9 3","pages":"4383-4393"},"PeriodicalIF":8.2,"publicationDate":"2024-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140820277","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-02-12DOI: 10.1109/TIV.2024.3364652
Yumeng Li;Yunhe Zhang;Tong Guo;Yu Liu;Yisheng Lv;Wenbo Du
The escalating density of airspace has led to sharply increased conflicts between aircraft. Efficient and scalable conflict resolution methods are crucial to mitigate collision risks. Existing learning-based methods become less effective as the scale of aircraft increases due to their redundant information representations. In this paper, to accommodate the increased airspace density, a novel graph reinforcement learning (GRL) method is presented to efficiently learn deconfliction strategies. A time-evolving conflict graph is exploited to represent the local state of individual aircraft and the global spatiotemporal relationships between them. Equipped with the conflict graph, GRL can efficiently learn deconfliction strategies by selectively aggregating aircraft state information through a multi-head attention-boosted graph neural network. Furthermore, a temporal regularization mechanism is proposed to enhance learning stability in highly dynamic environments. Comprehensive experimental studies have been conducted on an OpenAI Gym-based flight simulator. Compared with the existing state-of-the-art learning-based methods, the results demonstrate that GRL can save much training time while achieving significantly better deconfliction strategies in terms of safety and efficiency metrics. In addition, GRL has a strong power of scalability and robustness with increasing aircraft scale.
{"title":"Graph Reinforcement Learning for Multi-Aircraft Conflict Resolution","authors":"Yumeng Li;Yunhe Zhang;Tong Guo;Yu Liu;Yisheng Lv;Wenbo Du","doi":"10.1109/TIV.2024.3364652","DOIUrl":"https://doi.org/10.1109/TIV.2024.3364652","url":null,"abstract":"The escalating density of airspace has led to sharply increased conflicts between aircraft. Efficient and scalable conflict resolution methods are crucial to mitigate collision risks. Existing learning-based methods become less effective as the scale of aircraft increases due to their redundant information representations. In this paper, to accommodate the increased airspace density, a novel graph reinforcement learning (GRL) method is presented to efficiently learn deconfliction strategies. A time-evolving conflict graph is exploited to represent the local state of individual aircraft and the global spatiotemporal relationships between them. Equipped with the conflict graph, GRL can efficiently learn deconfliction strategies by selectively aggregating aircraft state information through a multi-head attention-boosted graph neural network. Furthermore, a temporal regularization mechanism is proposed to enhance learning stability in highly dynamic environments. Comprehensive experimental studies have been conducted on an OpenAI Gym-based flight simulator. Compared with the existing state-of-the-art learning-based methods, the results demonstrate that GRL can save much training time while achieving significantly better deconfliction strategies in terms of safety and efficiency metrics. In addition, GRL has a strong power of scalability and robustness with increasing aircraft scale.","PeriodicalId":36532,"journal":{"name":"IEEE Transactions on Intelligent Vehicles","volume":"9 3","pages":"4529-4540"},"PeriodicalIF":8.2,"publicationDate":"2024-02-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140820422","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-02-08DOI: 10.1109/TIV.2024.3363830
Lu Yang;Yiwen Bai;Fenglei Ren;Chongke Bi;Ronghui Zhang
Semantic segmentation is an important research topic in the environment perception of intelligent vehicles. Many semantic segmentation networks based on bilateral architecture have been proven effective. However, semantic segmentation networks based on this architecture has the risk of pixel classification errors and small objects being overwhelmed. In this paper, we solve the problem by proposing a novel three-branch architecture network called LCFNets. Compared to existing bilateral architecture, LCFNets introduce compensation branch for the first time to preserve the features of original images. Through two efficient modules, Lightweight Detail Guidance Fusion Module (L-DGF) and Lightweight Semantic Guidance Fusion Module (L-SGF), detail and semantic branches are allowed to selectively extract features from this branch. To balance the three-branch features and guide them to fuse effectively, a novel aggregation layer is designed. Depth-wise Convolution Pyramid Pooling module (DCPP) and Total Guidance Fusion Module (TGF) enable the aggregation layer to extract the global receptive field and realize multi-branch aggregation with fewer calculation complexity. Extensive experiments on Cityscapes and CamVid datasets have shown that our family of LCFNets provide a better trade-off between speed and accuracy. With the full resolution input and no ImageNet pre-training, LCFNet-slim achieves 76.86% mIoU at 114.36 FPS and LCFNet achieves 77.96% mIoU at 92.37 FPS on Cityscapes. On the other hand, LCFNet-super achieves 79.10% mIoU at 47.46 FPS.
{"title":"LCFNets: Compensation Strategy for Real-Time Semantic Segmentation of Autonomous Driving","authors":"Lu Yang;Yiwen Bai;Fenglei Ren;Chongke Bi;Ronghui Zhang","doi":"10.1109/TIV.2024.3363830","DOIUrl":"https://doi.org/10.1109/TIV.2024.3363830","url":null,"abstract":"Semantic segmentation is an important research topic in the environment perception of intelligent vehicles. Many semantic segmentation networks based on bilateral architecture have been proven effective. However, semantic segmentation networks based on this architecture has the risk of pixel classification errors and small objects being overwhelmed. In this paper, we solve the problem by proposing a novel three-branch architecture network called LCFNets. Compared to existing bilateral architecture, LCFNets introduce compensation branch for the first time to preserve the features of original images. Through two efficient modules, Lightweight Detail Guidance Fusion Module (L-DGF) and Lightweight Semantic Guidance Fusion Module (L-SGF), detail and semantic branches are allowed to selectively extract features from this branch. To balance the three-branch features and guide them to fuse effectively, a novel aggregation layer is designed. Depth-wise Convolution Pyramid Pooling module (DCPP) and Total Guidance Fusion Module (TGF) enable the aggregation layer to extract the global receptive field and realize multi-branch aggregation with fewer calculation complexity. Extensive experiments on Cityscapes and CamVid datasets have shown that our family of LCFNets provide a better trade-off between speed and accuracy. With the full resolution input and no ImageNet pre-training, LCFNet-slim achieves 76.86% mIoU at 114.36 FPS and LCFNet achieves 77.96% mIoU at 92.37 FPS on Cityscapes. On the other hand, LCFNet-super achieves 79.10% mIoU at 47.46 FPS.","PeriodicalId":36532,"journal":{"name":"IEEE Transactions on Intelligent Vehicles","volume":"9 4","pages":"4715-4729"},"PeriodicalIF":8.2,"publicationDate":"2024-02-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141315184","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Cooperative Driving Automation (CDA) stands at the forefront of the evolving landscape of vehicle automation, elevating driving capabilities within intricate real-world environments. This research aims to navigate the path toward the future of CDA by offering a thorough examination from the perspective of Planning and Control (PnC). It classifies state-of-the-art literature according to the CDA classes defined by the Society of Automotive Engineers (SAE). The strengths, weaknesses, and requirements of PnC for each CDA class are analyzed. This analysis helps identify areas that need improvement and provides insights into potential research directions. The research further discusses the evolution directions for CDA, providing valuable insights into the potential areas for further enhancement and enrichment of CDA research. The suggested areas include: Control robustness against disturbance; Risk-aware planning in a mixed environment of Connected and Automated Vehicles (CAVs) and Human-driven Vehicles (HVs); Vehicle-signal coupled modeling for coordination enhancement; Vehicle grouping to enhance the mobility of platooning.
{"title":"Towards the Next Level of Vehicle Automation Through Cooperative Driving: A Roadmap From Planning and Control Perspective","authors":"Haoran Wang;Yongwei Feng;Yonglin Tian;Ziran Wang;Jia Hu;Masayoshi Tomizuka","doi":"10.1109/TIV.2024.3363873","DOIUrl":"https://doi.org/10.1109/TIV.2024.3363873","url":null,"abstract":"Cooperative Driving Automation (CDA) stands at the forefront of the evolving landscape of vehicle automation, elevating driving capabilities within intricate real-world environments. This research aims to navigate the path toward the future of CDA by offering a thorough examination from the perspective of Planning and Control (PnC). It classifies state-of-the-art literature according to the CDA classes defined by the Society of Automotive Engineers (SAE). The strengths, weaknesses, and requirements of PnC for each CDA class are analyzed. This analysis helps identify areas that need improvement and provides insights into potential research directions. The research further discusses the evolution directions for CDA, providing valuable insights into the potential areas for further enhancement and enrichment of CDA research. The suggested areas include: Control robustness against disturbance; Risk-aware planning in a mixed environment of Connected and Automated Vehicles (CAVs) and Human-driven Vehicles (HVs); Vehicle-signal coupled modeling for coordination enhancement; Vehicle grouping to enhance the mobility of platooning.","PeriodicalId":36532,"journal":{"name":"IEEE Transactions on Intelligent Vehicles","volume":"9 3","pages":"4335-4347"},"PeriodicalIF":8.2,"publicationDate":"2024-02-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140820419","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-02-07DOI: 10.1109/TIV.2024.3363232
Xuan Li;Enlu Liu;Tianyu Shen;Jun Huang;Fei-Yue Wang
The latest developments in parallel driving foreshadow the possibility of delivering intelligence across organizations using foundation models. As is well-known, there are limitations in scenario acquisition in the field of intelligent vehicles (IV), such as efficiency, diversity, and complexity, which hinder in-depth research of vehicle intelligence. To address this issue, this manuscript draws inspiration from scenarios engineering, parallel driving and introduces a pioneering framework for scenario generation, leveraging the ChatGPT, denoted as SeGPT. Within this framework, we define a trajectory scenario and design prompts engineering to generate complex and challenging scenarios. Furthermore, SeGPT, in combination with “Three Modes”, foundation models, vehicle operating system, and other advanced infrastructure, holds the potential to achieve higher levels of autonomous driving. Experimental outcomes substantiate SeGPT's adeptness in producing a spectrum of varied scenarios, underscoring its potential to augment the development of trajectory prediction algorithms. These findings mark significant progress in the domain of scenario generation, also pointing towards new directions in the research of vehicle intelligence and scenarios engineering.
{"title":"ChatGPT-Based Scenario Engineer: A New Framework on Scenario Generation for Trajectory Prediction","authors":"Xuan Li;Enlu Liu;Tianyu Shen;Jun Huang;Fei-Yue Wang","doi":"10.1109/TIV.2024.3363232","DOIUrl":"https://doi.org/10.1109/TIV.2024.3363232","url":null,"abstract":"The latest developments in parallel driving foreshadow the possibility of delivering intelligence across organizations using foundation models. As is well-known, there are limitations in scenario acquisition in the field of intelligent vehicles (IV), such as efficiency, diversity, and complexity, which hinder in-depth research of vehicle intelligence. To address this issue, this manuscript draws inspiration from scenarios engineering, parallel driving and introduces a pioneering framework for scenario generation, leveraging the ChatGPT, denoted as SeGPT. Within this framework, we define a trajectory scenario and design prompts engineering to generate complex and challenging scenarios. Furthermore, SeGPT, in combination with “Three Modes”, foundation models, vehicle operating system, and other advanced infrastructure, holds the potential to achieve higher levels of autonomous driving. Experimental outcomes substantiate SeGPT's adeptness in producing a spectrum of varied scenarios, underscoring its potential to augment the development of trajectory prediction algorithms. These findings mark significant progress in the domain of scenario generation, also pointing towards new directions in the research of vehicle intelligence and scenarios engineering.","PeriodicalId":36532,"journal":{"name":"IEEE Transactions on Intelligent Vehicles","volume":"9 3","pages":"4422-4431"},"PeriodicalIF":8.2,"publicationDate":"2024-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140820329","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
We investigate the problem of autonomous racing among teams of cooperative agents that are subject to realistic racing rules. Our work extends previous research on hierarchical control in head-to-head autonomous racing by considering a generalized version of the problem while maintaining the two-level hierarchical control structure. A high-level tactical planner constructs a discrete game that encodes the complex rules using simplified dynamics to produce a sequence of target waypoints. The low-level path planner uses these waypoints as a reference trajectory and computes high-resolution control inputs by solving a simplified formulation of a racing game with a simplified representation of the realistic racing rules. We explore two approaches for the low-level path planner: training a multi-agent reinforcement learning (MARL) policy and solving a linear-quadratic Nash game (LQNG) approximation. We evaluate our controllers on simple and complex tracks against three baselines: an end-to-end MARL controller, a MARL controller tracking a fixed racing line, and an LQNG controller tracking a fixed racing line. Quantitative results show our hierarchical methods outperform the baselines in terms of race wins, overall team performance, and compliance with the rules. Qualitatively, we observe the hierarchical controllers mimic actions performed by expert human drivers such as coordinated overtaking, defending against multiple opponents, and long-term planning for delayed advantages.
{"title":"Hierarchical Control for Cooperative Teams in Competitive Autonomous Racing","authors":"Rishabh Saumil Thakkar;Aryaman Singh Samyal;David Fridovich-Keil;Zhe Xu;Ufuk Topcu","doi":"10.1109/TIV.2024.3363177","DOIUrl":"https://doi.org/10.1109/TIV.2024.3363177","url":null,"abstract":"We investigate the problem of autonomous racing among teams of cooperative agents that are subject to realistic racing rules. Our work extends previous research on hierarchical control in head-to-head autonomous racing by considering a generalized version of the problem while maintaining the two-level hierarchical control structure. A high-level tactical planner constructs a discrete game that encodes the complex rules using simplified dynamics to produce a sequence of target waypoints. The low-level path planner uses these waypoints as a reference trajectory and computes high-resolution control inputs by solving a simplified formulation of a racing game with a simplified representation of the realistic racing rules. We explore two approaches for the low-level path planner: training a multi-agent reinforcement learning (MARL) policy and solving a linear-quadratic Nash game (LQNG) approximation. We evaluate our controllers on simple and complex tracks against three baselines: an end-to-end MARL controller, a MARL controller tracking a fixed racing line, and an LQNG controller tracking a fixed racing line. Quantitative results show our hierarchical methods outperform the baselines in terms of race wins, overall team performance, and compliance with the rules. Qualitatively, we observe the hierarchical controllers mimic actions performed by expert human drivers such as coordinated overtaking, defending against multiple opponents, and long-term planning for delayed advantages.","PeriodicalId":36532,"journal":{"name":"IEEE Transactions on Intelligent Vehicles","volume":"9 5","pages":"4845-4860"},"PeriodicalIF":14.0,"publicationDate":"2024-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141964746","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-02-07DOI: 10.1109/TIV.2024.3360410
Gaurav Sharma;Rajesh Rajamani
Teleoperation is increasingly used in the operation of delivery robots and is beginning to be utilized for certain autonomous vehicle intervention applications. This paper addresses the challenges in teleoperation of an autonomous vehicle due to latencies in wireless communication between the remote vehicle and the teleoperator station. Camera video images and Lidar data are typically delayed during wireless transmission but are critical for proper display of the remote vehicle's real-time road environment to the teleoperator. Data collected with experiments in this project show that a 0.5 second delay in real-time display makes it extremely difficult for the teleoperator to control the remote vehicle. This problem is addressed in the paper by using a predictive display (PD) system which provides intermediate updates of the remote vehicle's environment while waiting for actual camera images. The predictive display utilizes estimated positions of the ego vehicle and of other vehicles on the road computed using model-based extended Kalman filters. A crucial result presented in the paper is that vehicle motion models need to be inertial rather than relative and so tracking of other vehicles requires accurate localization of the ego vehicle itself. An experimental study using 5 human teleoperators is conducted to compare teleoperation performance with and without predictive display. A 0.5 second time-delay in camera images makes it impossible to control the vehicle to stay in its lane on curved roads, but the use of the developed predictive display system enables safe remote vehicle control with almost as accurate a performance as the delay-free case.
{"title":"Teleoperation Enhancement for Autonomous Vehicles Using Estimation Based Predictive Display","authors":"Gaurav Sharma;Rajesh Rajamani","doi":"10.1109/TIV.2024.3360410","DOIUrl":"https://doi.org/10.1109/TIV.2024.3360410","url":null,"abstract":"Teleoperation is increasingly used in the operation of delivery robots and is beginning to be utilized for certain autonomous vehicle intervention applications. This paper addresses the challenges in teleoperation of an autonomous vehicle due to latencies in wireless communication between the remote vehicle and the teleoperator station. Camera video images and Lidar data are typically delayed during wireless transmission but are critical for proper display of the remote vehicle's real-time road environment to the teleoperator. Data collected with experiments in this project show that a 0.5 second delay in real-time display makes it extremely difficult for the teleoperator to control the remote vehicle. This problem is addressed in the paper by using a predictive display (PD) system which provides intermediate updates of the remote vehicle's environment while waiting for actual camera images. The predictive display utilizes estimated positions of the ego vehicle and of other vehicles on the road computed using model-based extended Kalman filters. A crucial result presented in the paper is that vehicle motion models need to be inertial rather than relative and so tracking of other vehicles requires accurate localization of the ego vehicle itself. An experimental study using 5 human teleoperators is conducted to compare teleoperation performance with and without predictive display. A 0.5 second time-delay in camera images makes it impossible to control the vehicle to stay in its lane on curved roads, but the use of the developed predictive display system enables safe remote vehicle control with almost as accurate a performance as the delay-free case.","PeriodicalId":36532,"journal":{"name":"IEEE Transactions on Intelligent Vehicles","volume":"9 3","pages":"4456-4469"},"PeriodicalIF":8.2,"publicationDate":"2024-02-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140820325","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This article investigates the parallel trajectory tracking control of fully-actuated maritime autonomous surface ships (MASSs) in the presence of multiple stationary/moving ocean obstacles. A safety-critical parallel control architecture is proposed for the trajectory tracking control of MASSs. Specifically, an artificial MASS system is constructed based on a data-driven learning predictor where real-time and historical navigation data are both utilized to achieve the estimation of the unknown weights of Taylor polynomials and Fourier series. Then, a parallel trajectory tracking control law is designed based on the artificial system such that the MASS is capable of track the reference trajectory positively. Finally, integral control barrier functions are employed to encode input and safety constraints. A safety optimization signal is augmented to the designed parallel control law to achieve the collision avoidance of all ocean obstacles. Based on the stability and safety analyses, the tracking errors of the actual MASS system are verified to be uniformly ultimately bounded and the MASS system is safe. Numerical examples confirm the effectiveness of the designed safety-critical parallel trajectory tracking control scheme for the MASS.
本文研究了在存在多个静止/移动海洋障碍物的情况下，全自动海上水面舰艇（MASSs）的并行轨迹跟踪控制。针对 MASS 的轨迹跟踪控制，提出了一种安全关键型并行控制架构。具体来说，基于数据驱动的学习预测器构建了一个人工 MASS 系统，利用实时和历史导航数据实现对泰勒多项式和傅里叶级数未知权重的估计。然后，基于人工系统设计并行轨迹跟踪控制法则，使 MASS 能够积极跟踪参考轨迹。最后，采用积分控制障碍函数对输入和安全约束进行编码。安全优化信号被添加到设计的并行控制法则中，以实现对所有海洋障碍物的碰撞规避。基于稳定性和安全性分析，验证了实际 MASS 系统的跟踪误差是均匀终界的，并且 MASS 系统是安全的。数值实例证实了所设计的 MASS 安全关键并行轨迹跟踪控制方案的有效性。
{"title":"Safety-Critical Parallel Trajectory Tracking Control of Maritime Autonomous Surface Ships Based on Integral Control Barrier Functions","authors":"Jiaxue Xu;Nan Gu;Dan Wang;Tieshan Li;Bing Han;Zhouhua Peng","doi":"10.1109/TIV.2024.3361477","DOIUrl":"https://doi.org/10.1109/TIV.2024.3361477","url":null,"abstract":"This article investigates the parallel trajectory tracking control of fully-actuated maritime autonomous surface ships (MASSs) in the presence of multiple stationary/moving ocean obstacles. A safety-critical parallel control architecture is proposed for the trajectory tracking control of MASSs. Specifically, an artificial MASS system is constructed based on a data-driven learning predictor where real-time and historical navigation data are both utilized to achieve the estimation of the unknown weights of Taylor polynomials and Fourier series. Then, a parallel trajectory tracking control law is designed based on the artificial system such that the MASS is capable of track the reference trajectory positively. Finally, integral control barrier functions are employed to encode input and safety constraints. A safety optimization signal is augmented to the designed parallel control law to achieve the collision avoidance of all ocean obstacles. Based on the stability and safety analyses, the tracking errors of the actual MASS system are verified to be uniformly ultimately bounded and the MASS system is safe. Numerical examples confirm the effectiveness of the designed safety-critical parallel trajectory tracking control scheme for the MASS.","PeriodicalId":36532,"journal":{"name":"IEEE Transactions on Intelligent Vehicles","volume":"9 5","pages":"4979-4988"},"PeriodicalIF":14.0,"publicationDate":"2024-02-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"141964760","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：