In this paper, we investigate the use of Vision Transformers for processing and understanding visual data in an autonomous driving setting. Specifically, we explore the use of Vision Transformers for semantic segmentation and monocular depth estimation using only a single image as input. We present state-of-the-art Vision Transformers for these tasks and combine them into a multitask model. Through multiple experiments on four different street image datasets, we demonstrate that the multitask approach significantly reduces inference time while maintaining high accuracy for both tasks. Additionally, we show that changing the size of the Transformer-based backbone can be used as a trade-off between inference speed and accuracy. Furthermore, we investigate the use of synthetic data for pre-training and show that it effectively increases the accuracy of the model when real-world data is limited.
{"title":"A Multi-Task Vision Transformer for Segmentation and Monocular Depth Estimation for Autonomous Vehicles","authors":"Durga Prasad Bavirisetti;Herman Ryen Martinsen;Gabriel Hanssen Kiss;Frank Lindseth","doi":"10.1109/OJITS.2023.3335648","DOIUrl":"10.1109/OJITS.2023.3335648","url":null,"abstract":"In this paper, we investigate the use of Vision Transformers for processing and understanding visual data in an autonomous driving setting. Specifically, we explore the use of Vision Transformers for semantic segmentation and monocular depth estimation using only a single image as input. We present state-of-the-art Vision Transformers for these tasks and combine them into a multitask model. Through multiple experiments on four different street image datasets, we demonstrate that the multitask approach significantly reduces inference time while maintaining high accuracy for both tasks. Additionally, we show that changing the size of the Transformer-based backbone can be used as a trade-off between inference speed and accuracy. Furthermore, we investigate the use of synthetic data for pre-training and show that it effectively increases the accuracy of the model when real-world data is limited.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"4 ","pages":"909-928"},"PeriodicalIF":0.0,"publicationDate":"2023-11-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10330677","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"138576854","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-11-28DOI: 10.1109/OJITS.2023.3335303
Pushkin Kachroo;Shaurya Agarwal;Animesh Biswas;Archie J. Huang
Nonlocal calculus-based macroscopic traffic models overcome the limitations of classical local models in accurately capturing traffic flow dynamics. These models incorporate “nonlocal” elements by considering the speed as a weighted mean of downstream traffic density, aligning it more closely with realistic driving behaviors. The primary contributions of this research are manifold. Firstly, we choose a nonlocal LWR model and Greenshields fundamental diagram and prove that this traffic flow model satisfies the well-posed conditions. Furthermore, we prove that the chosen model maintains bounded states, laying the groundwork for developing numerically stable schemes. Subsequently, the efficacy of the proposed nonlocal model is evaluated through extensive field validation using real traffic data from the NGSIM dataset and developing a stable numerical scheme. These validation results highlight the superiority of the nonlocal model in capturing traffic characteristics compared to its local counterpart and establish its enhanced accuracy in reproducing complex traffic behavior. Therefore, this research expands both the theoretical constructs within the field and substantiates its practical applicability.
{"title":"Nonlocal Calculus-Based Macroscopic Traffic Model: Development, Analysis, and Validation","authors":"Pushkin Kachroo;Shaurya Agarwal;Animesh Biswas;Archie J. Huang","doi":"10.1109/OJITS.2023.3335303","DOIUrl":"https://doi.org/10.1109/OJITS.2023.3335303","url":null,"abstract":"Nonlocal calculus-based macroscopic traffic models overcome the limitations of classical local models in accurately capturing traffic flow dynamics. These models incorporate “nonlocal” elements by considering the speed as a weighted mean of downstream traffic density, aligning it more closely with realistic driving behaviors. The primary contributions of this research are manifold. Firstly, we choose a nonlocal LWR model and Greenshields fundamental diagram and prove that this traffic flow model satisfies the well-posed conditions. Furthermore, we prove that the chosen model maintains bounded states, laying the groundwork for developing numerically stable schemes. Subsequently, the efficacy of the proposed nonlocal model is evaluated through extensive field validation using real traffic data from the NGSIM dataset and developing a stable numerical scheme. These validation results highlight the superiority of the nonlocal model in capturing traffic characteristics compared to its local counterpart and establish its enhanced accuracy in reproducing complex traffic behavior. Therefore, this research expands both the theoretical constructs within the field and substantiates its practical applicability.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"4 ","pages":"900-908"},"PeriodicalIF":0.0,"publicationDate":"2023-11-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10330738","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"138558041","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-11-24DOI: 10.1109/OJITS.2023.3336464
Rainer Trauth;Korbinian Moller;Johannes Betz
Autonomous vehicles face numerous challenges to ensure safe operation in unpredictable and hazardous conditions. The autonomous driving environment is characterized by high uncertainty, especially in occluded areas with limited information about the surrounding obstacles. This work aims to provide a trajectory planner to solve these unsafe environments. The work proposes an approach combining a visibility model, contextual environmental information, and behavioral planning algorithms to predict the likelihood of occlusions and collision probabilities. Ultimately, this allows us to estimate the potential harm from collisions with pedestrians in occluded situations. The primary goal of our proposed approach is to minimize the risk of hitting pedestrians and to establish a predefined, adjustable maximum level of harm. We show several practical applications for informing a sampling-based trajectory planner about occluded areas to increase safety. In addition, to respond to possible high-risk situations, we introduce an adjustable threshold that governs the vehicle’s speed when encountering uncertain situations and strategies to maximize the vehicle’s visible area. In implementing our novel methodology, we analyzed several real-world scenarios in a simulation environment. Our results indicate that combining occlusion-aware trajectory planning algorithms and harm estimation significantly influences vehicle driving behavior, especially in risky situations. The code used in this research is publicly available as open-source software and can be accessed at the following link: https://github.com/TUM-AVS/Frenetix-Motion-Planner .
{"title":"Toward Safer Autonomous Vehicles: Occlusion-Aware Trajectory Planning to Minimize Risky Behavior","authors":"Rainer Trauth;Korbinian Moller;Johannes Betz","doi":"10.1109/OJITS.2023.3336464","DOIUrl":"https://doi.org/10.1109/OJITS.2023.3336464","url":null,"abstract":"Autonomous vehicles face numerous challenges to ensure safe operation in unpredictable and hazardous conditions. The autonomous driving environment is characterized by high uncertainty, especially in occluded areas with limited information about the surrounding obstacles. This work aims to provide a trajectory planner to solve these unsafe environments. The work proposes an approach combining a visibility model, contextual environmental information, and behavioral planning algorithms to predict the likelihood of occlusions and collision probabilities. Ultimately, this allows us to estimate the potential harm from collisions with pedestrians in occluded situations. The primary goal of our proposed approach is to minimize the risk of hitting pedestrians and to establish a predefined, adjustable maximum level of harm. We show several practical applications for informing a sampling-based trajectory planner about occluded areas to increase safety. In addition, to respond to possible high-risk situations, we introduce an adjustable threshold that governs the vehicle’s speed when encountering uncertain situations and strategies to maximize the vehicle’s visible area. In implementing our novel methodology, we analyzed several real-world scenarios in a simulation environment. Our results indicate that combining occlusion-aware trajectory planning algorithms and harm estimation significantly influences vehicle driving behavior, especially in risky situations. The code used in this research is publicly available as open-source software and can be accessed at the following link: \u0000<uri>https://github.com/TUM-AVS/Frenetix-Motion-Planner</uri>\u0000.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"4 ","pages":"929-942"},"PeriodicalIF":0.0,"publicationDate":"2023-11-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10328654","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"138633904","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-11-23DOI: 10.1109/OJITS.2023.3335817
Younghoon Seo;Jihyeok Park;Gyungtaek Oh;Hyungjoo Kim;Jia Hu;Jaehyun So
The unstructured-textual crash descriptions recorded by police officers is rarely utilized, despite containing detailed information on traffic situations. This lack of utilization is mainly due to the difficulty in analyzing text data, as there is currently no innovative methodology for extracting meaningful information from it. Given limitations and challenges in analyzing traffic crash descriptions, this study developed a methodology to classify significant words in unstructured data that describe traffic crash scenarios into standardized data. Ultimately, a natural language processing technique, specifically a bidirectional encoder representation from transformer (BERT), was used to extract meaningful information from crash descriptions. This BERT-based model effectively extracts information on the exact collision point and the pre-crash vehicle maneuver from crash descriptions. Its practical approach allows for the interpretation of traffic crash descriptions and outperforms other natural language processing models. Importantly, this method of extracting crash scene information from traffic crash descriptions can aid in better comprehending the unique characteristics of traffic crashes. This comprehension can ultimately aid in the development of appropriate countermeasures, leading to the prevention of future traffic crashes.
{"title":"Text Classification Modeling Approach on Imbalanced-Unstructured Traffic Accident Descriptions Data","authors":"Younghoon Seo;Jihyeok Park;Gyungtaek Oh;Hyungjoo Kim;Jia Hu;Jaehyun So","doi":"10.1109/OJITS.2023.3335817","DOIUrl":"https://doi.org/10.1109/OJITS.2023.3335817","url":null,"abstract":"The unstructured-textual crash descriptions recorded by police officers is rarely utilized, despite containing detailed information on traffic situations. This lack of utilization is mainly due to the difficulty in analyzing text data, as there is currently no innovative methodology for extracting meaningful information from it. Given limitations and challenges in analyzing traffic crash descriptions, this study developed a methodology to classify significant words in unstructured data that describe traffic crash scenarios into standardized data. Ultimately, a natural language processing technique, specifically a bidirectional encoder representation from transformer (BERT), was used to extract meaningful information from crash descriptions. This BERT-based model effectively extracts information on the exact collision point and the pre-crash vehicle maneuver from crash descriptions. Its practical approach allows for the interpretation of traffic crash descriptions and outperforms other natural language processing models. Importantly, this method of extracting crash scene information from traffic crash descriptions can aid in better comprehending the unique characteristics of traffic crashes. This comprehension can ultimately aid in the development of appropriate countermeasures, leading to the prevention of future traffic crashes.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"4 ","pages":"955-965"},"PeriodicalIF":0.0,"publicationDate":"2023-11-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10327780","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"138739588","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Bicycles are frequently involved in traffic collisions with vehicles, particularly when sudden changes in direction occur. This paper presents a robust risk-predictive braking policy to ensure collision avoidance in all possible crossing behaviors of a bicycle. The policy controls the vehicle to follow an upper limit of the safe speed before the bicycle changes direction, ensuring that the vehicle can stop in time by the advanced emergency braking system before a collision occurs in any situation. The upper limit of the safe speed is the solution of an intractable robust optimization problem. Therefore, a scenario approach is adapted to develop a tractable approximate problem for the original robust optimization problem. The feasibility and optimality of the problem reduction are theoretically proved. A bisection method-based fast algorithm is designed to solve the approximate problem of the original robust optimization problem, making it applicable in practical scenarios. The convergence of the algorithm is also proven. The effectiveness of the proposed method is validated through hardware-in-the-loop simulations using CarMaker.
{"title":"Robust Optimal Braking Policy for Avoiding Collision With Front Bicycle","authors":"Xun Shen;Yan Zhang;Xingguo Zhang;Pongsathorn Raksincharoensak;Kazumune Hashimoto","doi":"10.1109/OJITS.2023.3335397","DOIUrl":"https://doi.org/10.1109/OJITS.2023.3335397","url":null,"abstract":"Bicycles are frequently involved in traffic collisions with vehicles, particularly when sudden changes in direction occur. This paper presents a robust risk-predictive braking policy to ensure collision avoidance in all possible crossing behaviors of a bicycle. The policy controls the vehicle to follow an upper limit of the safe speed before the bicycle changes direction, ensuring that the vehicle can stop in time by the advanced emergency braking system before a collision occurs in any situation. The upper limit of the safe speed is the solution of an intractable robust optimization problem. Therefore, a scenario approach is adapted to develop a tractable approximate problem for the original robust optimization problem. The feasibility and optimality of the problem reduction are theoretically proved. A bisection method-based fast algorithm is designed to solve the approximate problem of the original robust optimization problem, making it applicable in practical scenarios. The convergence of the algorithm is also proven. The effectiveness of the proposed method is validated through hardware-in-the-loop simulations using CarMaker.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"4 ","pages":"943-954"},"PeriodicalIF":0.0,"publicationDate":"2023-11-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10325675","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"138633918","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-11-21DOI: 10.1109/OJITS.2023.3334393
Julian Teusch;Jan Niklas Gremmel;Christian Koetsier;Fatema Tuj Johora;Monika Sester;David M. Woisetschläger;Jörg P. Müller
Shared mobility has emerged as a sustainable alternative to both private transportation and traditional public transport, promising to reduce the number of private vehicles on roads while offering users greater flexibility. Today, urban areas are home to a myriad of innovative services, including car-sharing, ride-sharing, and micromobility solutions like moped-sharing, bike-sharing, and e-scooter-sharing. Given the intense competition and the inherent operational complexities of shared mobility systems, providers are increasingly seeking specialized decision-support methodologies to boost operational efficiency. While recent research indicates that advanced machine learning methods can tackle the intricate challenges in shared mobility management decisions, a thorough evaluation of existing research is essential to fully grasp its potential and pinpoint areas needing further exploration. This paper presents a systematic literature review that specifically targets the application of Machine Learning for decision-making in Shared Mobility Systems. Our review underscores that Machine Learning offers methodological solutions to specific management challenges crucial for the effective operation of Shared Mobility Systems. We delve into the methods and datasets employed, spotlight research trends, and pinpoint research gaps. Our findings culminate in a comprehensive framework of Machine Learning techniques designed to bolster managerial decision-making in addressing challenges specific to Shared Mobility across various levels.
{"title":"A Systematic Literature Review on Machine Learning in Shared Mobility","authors":"Julian Teusch;Jan Niklas Gremmel;Christian Koetsier;Fatema Tuj Johora;Monika Sester;David M. Woisetschläger;Jörg P. Müller","doi":"10.1109/OJITS.2023.3334393","DOIUrl":"https://doi.org/10.1109/OJITS.2023.3334393","url":null,"abstract":"Shared mobility has emerged as a sustainable alternative to both private transportation and traditional public transport, promising to reduce the number of private vehicles on roads while offering users greater flexibility. Today, urban areas are home to a myriad of innovative services, including car-sharing, ride-sharing, and micromobility solutions like moped-sharing, bike-sharing, and e-scooter-sharing. Given the intense competition and the inherent operational complexities of shared mobility systems, providers are increasingly seeking specialized decision-support methodologies to boost operational efficiency. While recent research indicates that advanced machine learning methods can tackle the intricate challenges in shared mobility management decisions, a thorough evaluation of existing research is essential to fully grasp its potential and pinpoint areas needing further exploration. This paper presents a systematic literature review that specifically targets the application of Machine Learning for decision-making in Shared Mobility Systems. Our review underscores that Machine Learning offers methodological solutions to specific management challenges crucial for the effective operation of Shared Mobility Systems. We delve into the methods and datasets employed, spotlight research trends, and pinpoint research gaps. Our findings culminate in a comprehensive framework of Machine Learning techniques designed to bolster managerial decision-making in addressing challenges specific to Shared Mobility across various levels.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"4 ","pages":"870-899"},"PeriodicalIF":0.0,"publicationDate":"2023-11-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10325362","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"138550272","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Behind many traffic accidents, there are more frequent minor incidents (risky traffic situations) that may lead to severe accidents. Analyzing such minor incidents effectively reduces accidents, but the challenge is to design a method to collect and analyze such incident information. In this paper, we propose a novel platform that aggregates behavioral data from pedestrians and drivers using their smartphones and recognizes risky traffic situations from the aggregated data. We design a two-stage approach where the smartphones of pedestrians and vehicles act as local anomaly detectors for triggering the event detector and classifier in the post-stage at the cloud server to suppress the processing and communication overhead. We also introduce an unsupervised learning system to cope with unseen risky situations enabled by joint utilization of the autoencoder-based anomaly detector and the risky situation classifier. The evaluation is conducted through both simulation and real experiments. The simulation result shows the risky situation detector achieves an F-measure of 0.89. We also collected real data at a car driving course to evaluate the risky situation classifier. From the results, we have confirmed that the proposed method succeeded in classifying three risky traffic situations involving pedestrians and/or vehicles with an accuracy of 89.3%.
{"title":"Risky Traffic Situation Detection and Classification Using Smartphones","authors":"Akira Uchiyama;Akihito Hiromori;Ryota Akikawa;Hirozumi Yamaguchi;Teruo Higashino;Masaki Suzuki;Yasuhiko Hiehata;Takeshi Kitahara","doi":"10.1109/OJITS.2023.3333263","DOIUrl":"10.1109/OJITS.2023.3333263","url":null,"abstract":"Behind many traffic accidents, there are more frequent minor incidents (risky traffic situations) that may lead to severe accidents. Analyzing such minor incidents effectively reduces accidents, but the challenge is to design a method to collect and analyze such incident information. In this paper, we propose a novel platform that aggregates behavioral data from pedestrians and drivers using their smartphones and recognizes risky traffic situations from the aggregated data. We design a two-stage approach where the smartphones of pedestrians and vehicles act as local anomaly detectors for triggering the event detector and classifier in the post-stage at the cloud server to suppress the processing and communication overhead. We also introduce an unsupervised learning system to cope with unseen risky situations enabled by joint utilization of the autoencoder-based anomaly detector and the risky situation classifier. The evaluation is conducted through both simulation and real experiments. The simulation result shows the risky situation detector achieves an F-measure of 0.89. We also collected real data at a car driving course to evaluate the risky situation classifier. From the results, we have confirmed that the proposed method succeeded in classifying three risky traffic situations involving pedestrians and/or vehicles with an accuracy of 89.3%.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"4 ","pages":"846-857"},"PeriodicalIF":0.0,"publicationDate":"2023-11-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10318156","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135709423","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-11-13DOI: 10.1109/OJITS.2023.3332043
Madhumitha Sakthi;Marius Arvinte;Haris Vikalo
In recent years, automotive radar has attracted considerable attention due to the growing interest in autonomous driving technologies. Acquiring situational awareness using multimodal data collected at high sampling rates by various sensing devices including cameras, LiDAR, and radar requires considerable power, memory and compute resources which are often limited at an edge device. In this paper, we present a novel adaptive radar sub-sampling algorithm designed to identify regions that require more detailed/accurate reconstruction based on the information about prior environmental conditions, enabling near-optimal performance at considerably lower effective sampling rates. Designed to robustly perform under variable weather conditions, the algorithm was shown on the Oxford radar dataset to achieve accurate scene reconstruction utilizing only 10% of the collected samples in good weather. In the case of the RADIATE dataset acquired during extreme weather conditions (snow, fog), only 20% of the samples were sufficient to enable robust scene reconstruction. A further modification of the algorithm incorporates object motion to enable reliable identification of regions that require attention. This includes monitoring possible future occlusions caused by the objects detected in the present frame. Finally, we train a YOLO network on the RADIATE dataset to perform object detection, obtaining 6.6% AP50 improvement over the baseline Faster R-CNN network.
{"title":"Automotive Radar Sub-Sampling via Object Detection Networks: Leveraging Prior Signal Information","authors":"Madhumitha Sakthi;Marius Arvinte;Haris Vikalo","doi":"10.1109/OJITS.2023.3332043","DOIUrl":"10.1109/OJITS.2023.3332043","url":null,"abstract":"In recent years, automotive radar has attracted considerable attention due to the growing interest in autonomous driving technologies. Acquiring situational awareness using multimodal data collected at high sampling rates by various sensing devices including cameras, LiDAR, and radar requires considerable power, memory and compute resources which are often limited at an edge device. In this paper, we present a novel adaptive radar sub-sampling algorithm designed to identify regions that require more detailed/accurate reconstruction based on the information about prior environmental conditions, enabling near-optimal performance at considerably lower effective sampling rates. Designed to robustly perform under variable weather conditions, the algorithm was shown on the Oxford radar dataset to achieve accurate scene reconstruction utilizing only 10% of the collected samples in good weather. In the case of the RADIATE dataset acquired during extreme weather conditions (snow, fog), only 20% of the samples were sufficient to enable robust scene reconstruction. A further modification of the algorithm incorporates object motion to enable reliable identification of regions that require attention. This includes monitoring possible future occlusions caused by the objects detected in the present frame. Finally, we train a YOLO network on the RADIATE dataset to perform object detection, obtaining 6.6% AP50 improvement over the baseline Faster R-CNN network.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"4 ","pages":"858-869"},"PeriodicalIF":0.0,"publicationDate":"2023-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10315142","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135660176","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
A number of deep reinforcement-learning (RL) approaches propose to control traffic signals. Compared to traditional approaches, RL approaches can learn from higher-dimensionality input road and vehicle sensors and better adapt to varying traffic conditions resulting in reduced travel times (in simulation). However, these RL methods require training from massive traffic sensor data. To offset this relative inefficiency, some recent RL methods have the ability to first learn from small-scale networks and then generalize to unseen city-scale networks without additional retraining (zero-shot transfer). In this work, we study the robustness of such methods along two axes. First, sensor failures and GPS occlusions create missing-data challenges and we show that recent methods remain brittle in the face of these missing data. Second, we provide a more systematic study of the generalization ability of RL methods to new networks with different traffic regimes. Again, we identify the limitations of recent approaches. We then propose using a combination of distributional and vanilla reinforcement learning through a policy ensemble. Building upon the state-of-the-art previous model which uses a decentralized approach for large-scale traffic signal control with graph convolutional networks (GCNs), we first learn models using a distributional reinforcement learning (DisRL) approach. In particular, we use implicit quantile networks (IQN) to model the state-action return distribution with quantile regression. For traffic signal control problems, an ensemble of standard RL and DisRL yields superior performance across different scenarios, including different levels of missing sensor data and traffic flow patterns. Furthermore, the learning scheme of the resulting model can improve zero-shot transferability to different road network structures, including both synthetic networks and real-world networks (e.g., Luxembourg, Manhattan). We conduct extensive experiments to compare our approach to multi-agent reinforcement learning and traditional transportation approaches. Results show that the proposed method improves robustness and generalizability in the face of missing data, varying road networks, and traffic flows.
{"title":"Improving the Generalizability and Robustness of Large-Scale Traffic Signal Control","authors":"Tianyu Shi;François-Xavier Devailly;Denis Larocque;Laurent Charlin","doi":"10.1109/OJITS.2023.3331689","DOIUrl":"10.1109/OJITS.2023.3331689","url":null,"abstract":"A number of deep reinforcement-learning (RL) approaches propose to control traffic signals. Compared to traditional approaches, RL approaches can learn from higher-dimensionality input road and vehicle sensors and better adapt to varying traffic conditions resulting in reduced travel times (in simulation). However, these RL methods require training from massive traffic sensor data. To offset this relative inefficiency, some recent RL methods have the ability to first learn from small-scale networks and then generalize to unseen city-scale networks without additional retraining (zero-shot transfer). In this work, we study the robustness of such methods along two axes. First, sensor failures and GPS occlusions create missing-data challenges and we show that recent methods remain brittle in the face of these missing data. Second, we provide a more systematic study of the generalization ability of RL methods to new networks with different traffic regimes. Again, we identify the limitations of recent approaches. We then propose using a combination of distributional and vanilla reinforcement learning through a policy ensemble. Building upon the state-of-the-art previous model which uses a decentralized approach for large-scale traffic signal control with graph convolutional networks (GCNs), we first learn models using a distributional reinforcement learning (DisRL) approach. In particular, we use implicit quantile networks (IQN) to model the state-action return distribution with quantile regression. For traffic signal control problems, an ensemble of standard RL and DisRL yields superior performance across different scenarios, including different levels of missing sensor data and traffic flow patterns. Furthermore, the learning scheme of the resulting model can improve zero-shot transferability to different road network structures, including both synthetic networks and real-world networks (e.g., Luxembourg, Manhattan). We conduct extensive experiments to compare our approach to multi-agent reinforcement learning and traditional transportation approaches. Results show that the proposed method improves robustness and generalizability in the face of missing data, varying road networks, and traffic flows.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"5 ","pages":"2-15"},"PeriodicalIF":0.0,"publicationDate":"2023-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10315958","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135610919","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-11-09DOI: 10.1109/OJITS.2023.3331447
Federico Gallo;Nicola Sacco;Francesco Corman
This paper addresses the problem of predicting the occupancy of urban public transport vehicles with a network-wide framework where the effects of the interactions between multiple lines are jointly considered. In particular, we propose and compare several occupancy predictors, each of them differing in the amount of information used and in the prediction model adopted. We consider two prediction models: a behavioral model that assumes an explicit relation between some observed variables and the occupancy, and a machine learning model based on the LightGBM algorithm. We evaluate the proposed network-wide prediction framework on two real-world case studies related to the public transport network of the Swiss city of Zurich. The results show that predicting the occupancy for a target line while simultaneously considering the other lines in the network allows significant improvements in the accuracy of the predictions, especially in the corridors served by different interacting lines. The described methodology could be used by public transport agencies to improve the accuracy of the crowding information provided to passengers and to increase the attractiveness of public transport systems.
{"title":"Network-Wide Public Transport Occupancy Prediction Framework With Multiple Line Interactions","authors":"Federico Gallo;Nicola Sacco;Francesco Corman","doi":"10.1109/OJITS.2023.3331447","DOIUrl":"10.1109/OJITS.2023.3331447","url":null,"abstract":"This paper addresses the problem of predicting the occupancy of urban public transport vehicles with a network-wide framework where the effects of the interactions between multiple lines are jointly considered. In particular, we propose and compare several occupancy predictors, each of them differing in the amount of information used and in the prediction model adopted. We consider two prediction models: a behavioral model that assumes an explicit relation between some observed variables and the occupancy, and a machine learning model based on the LightGBM algorithm. We evaluate the proposed network-wide prediction framework on two real-world case studies related to the public transport network of the Swiss city of Zurich. The results show that predicting the occupancy for a target line while simultaneously considering the other lines in the network allows significant improvements in the accuracy of the predictions, especially in the corridors served by different interacting lines. The described methodology could be used by public transport agencies to improve the accuracy of the crowding information provided to passengers and to increase the attractiveness of public transport systems.","PeriodicalId":100631,"journal":{"name":"IEEE Open Journal of Intelligent Transportation Systems","volume":"4 ","pages":"815-832"},"PeriodicalIF":0.0,"publicationDate":"2023-11-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10314412","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"135561163","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","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：