Pub Date : 2019-10-01DOI: 10.1109/ITSC.2019.8917027
G. Khodabandelou, Mehdi Katranji, Sami Kraiem, W. Kheriji, F. Hadj-Selem
With urge of demands on efficient transport planning policies along with surge of travel flow volumes due to fast urbanization, traffic speed forecasting becomes a canonical and thriving research domain. Furthermore, the vehicles speed plays a critical role in the level of congestion. Traffic speed estimation then helps transport authorities as well as network users to handle congestion over road infrastructures or at least provides a global picture of daily passenger flow. In this work, we propose the first methodology to forecast the future traffic speed over the road segments (i.e. links) exclusively based on traffic flow data using floating car data. For this study, we pre-process over one million vehicles flow for several network links spread all over the Greater Paris. A attention-based recurrent neural network is used to capture the correlation between the temporal sequences of traffic flow and that of speed. The attention layer learns patterns from weights of near-term traffic flow, thus extracts the inherent interdependency of traffic speed to many factors (e.g. incidents, rush hour, land use, etc.) in non-free-flow conditions. The results demonstrate the efficiency of the proposed model in traffic speed forecasting excluding additional data such as historic traffic speed and network graph contrary to cutting-edge work in the field. This is a substantial property since it allows avoiding the cumbersomeness in data mixing and facilitating resource availability.
{"title":"Attention-based Gated Recurrent Unit for Links Traffic Speed Forecasting","authors":"G. Khodabandelou, Mehdi Katranji, Sami Kraiem, W. Kheriji, F. Hadj-Selem","doi":"10.1109/ITSC.2019.8917027","DOIUrl":"https://doi.org/10.1109/ITSC.2019.8917027","url":null,"abstract":"With urge of demands on efficient transport planning policies along with surge of travel flow volumes due to fast urbanization, traffic speed forecasting becomes a canonical and thriving research domain. Furthermore, the vehicles speed plays a critical role in the level of congestion. Traffic speed estimation then helps transport authorities as well as network users to handle congestion over road infrastructures or at least provides a global picture of daily passenger flow. In this work, we propose the first methodology to forecast the future traffic speed over the road segments (i.e. links) exclusively based on traffic flow data using floating car data. For this study, we pre-process over one million vehicles flow for several network links spread all over the Greater Paris. A attention-based recurrent neural network is used to capture the correlation between the temporal sequences of traffic flow and that of speed. The attention layer learns patterns from weights of near-term traffic flow, thus extracts the inherent interdependency of traffic speed to many factors (e.g. incidents, rush hour, land use, etc.) in non-free-flow conditions. The results demonstrate the efficiency of the proposed model in traffic speed forecasting excluding additional data such as historic traffic speed and network graph contrary to cutting-edge work in the field. This is a substantial property since it allows avoiding the cumbersomeness in data mixing and facilitating resource availability.","PeriodicalId":6717,"journal":{"name":"2019 IEEE Intelligent Transportation Systems Conference (ITSC)","volume":"16 1","pages":"2577-2583"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"79146427","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-10-01DOI: 10.1109/ITSC.2019.8916970
Ruoyu Sun, Shaochi Hu, Huijing Zhao, M. Moze, F. Aioun, F. Guillemard
Autonomous driving is one of the current cutting edge technologies. For autonomous cars, their driving actions and trajectories should not only achieve autonomy and safety, but also obey human drivers’ behavior patterns, when sharing the roads with other human drivers on the highway. Traditional methods, though robust and interpretable, demands much human labor in engineering the complex mapping from current driving situation to vehicle’s future control. For newly developed deep-learning methods, though they can automatically learn such complex mapping from data and demands fewer humans’ engineering, they mostly act like black-box, and are less interpretable. We proposed a new combined method based on inverse reinforcement learning to harness the advantages of both. Experimental validations on lane-change prediction and human-like trajectory planning show that the proposed method approximates the state-of-the-art performance in modeling human trajectories, and is both interpretable and data-driven.
{"title":"Human-like Highway Trajectory Modeling based on Inverse Reinforcement Learning","authors":"Ruoyu Sun, Shaochi Hu, Huijing Zhao, M. Moze, F. Aioun, F. Guillemard","doi":"10.1109/ITSC.2019.8916970","DOIUrl":"https://doi.org/10.1109/ITSC.2019.8916970","url":null,"abstract":"Autonomous driving is one of the current cutting edge technologies. For autonomous cars, their driving actions and trajectories should not only achieve autonomy and safety, but also obey human drivers’ behavior patterns, when sharing the roads with other human drivers on the highway. Traditional methods, though robust and interpretable, demands much human labor in engineering the complex mapping from current driving situation to vehicle’s future control. For newly developed deep-learning methods, though they can automatically learn such complex mapping from data and demands fewer humans’ engineering, they mostly act like black-box, and are less interpretable. We proposed a new combined method based on inverse reinforcement learning to harness the advantages of both. Experimental validations on lane-change prediction and human-like trajectory planning show that the proposed method approximates the state-of-the-art performance in modeling human trajectories, and is both interpretable and data-driven.","PeriodicalId":6717,"journal":{"name":"2019 IEEE Intelligent Transportation Systems Conference (ITSC)","volume":"25 1","pages":"1482-1489"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81221363","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-10-01DOI: 10.1109/ITSC.2019.8917094
L. Hua, Wei Zheng, Shigen Gao
Learning and getting more information from past accident records to understand the accidents deeply are important to prevent future accidents. Most Chinese railway accidents are recorded in the form of text reports and the information about text reports is often underutilized due to the lack of effective mining and analysis tools. In this study, text mining and natural language process (NLP) techniques were used to analyze railway accident reports. More specifically, the multichannel convolutional neural network (M-CNN) and conditional random field (CRF) model were designed to extract accident risk factors. The experimental results shows that our system achieves good performance and can effectively extract risk factors from the accident reports. At the same time, the main risk factors leading to accidents are summarized from four aspects. The system can be used to solve problem areas and strengthen the safety management of the railway industry.
{"title":"Extraction and Analysis of Risk Factors from Chinese Railway Accident Reports","authors":"L. Hua, Wei Zheng, Shigen Gao","doi":"10.1109/ITSC.2019.8917094","DOIUrl":"https://doi.org/10.1109/ITSC.2019.8917094","url":null,"abstract":"Learning and getting more information from past accident records to understand the accidents deeply are important to prevent future accidents. Most Chinese railway accidents are recorded in the form of text reports and the information about text reports is often underutilized due to the lack of effective mining and analysis tools. In this study, text mining and natural language process (NLP) techniques were used to analyze railway accident reports. More specifically, the multichannel convolutional neural network (M-CNN) and conditional random field (CRF) model were designed to extract accident risk factors. The experimental results shows that our system achieves good performance and can effectively extract risk factors from the accident reports. At the same time, the main risk factors leading to accidents are summarized from four aspects. The system can be used to solve problem areas and strengthen the safety management of the railway industry.","PeriodicalId":6717,"journal":{"name":"2019 IEEE Intelligent Transportation Systems Conference (ITSC)","volume":"924 1","pages":"869-874"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"77539320","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-10-01DOI: 10.1109/ITSC.2019.8917197
Florian Geissler, Ralf Graefe
We present a multi-stage optimization method for efficient sensor deployment in traffic surveillance scenarios. Based on a genetic optimization scheme, our algorithm places an optimal number of roadside sensors to obtain full road coverage in the presence of obstacles and dynamic occlusions. The efficiency of the procedure is demonstrated for selected, realistic road sections. Our analysis helps to leverage the economic feasibility of distributed infrastructure sensor networks with high perception quality.
{"title":"Optimized sensor placement for dependable roadside infrastructures","authors":"Florian Geissler, Ralf Graefe","doi":"10.1109/ITSC.2019.8917197","DOIUrl":"https://doi.org/10.1109/ITSC.2019.8917197","url":null,"abstract":"We present a multi-stage optimization method for efficient sensor deployment in traffic surveillance scenarios. Based on a genetic optimization scheme, our algorithm places an optimal number of roadside sensors to obtain full road coverage in the presence of obstacles and dynamic occlusions. The efficiency of the procedure is demonstrated for selected, realistic road sections. Our analysis helps to leverage the economic feasibility of distributed infrastructure sensor networks with high perception quality.","PeriodicalId":6717,"journal":{"name":"2019 IEEE Intelligent Transportation Systems Conference (ITSC)","volume":"15 6","pages":"2408-2413"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"91427874","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-10-01DOI: 10.1109/ITSC.2019.8917282
Ming Chai, Haifeng Wang, Hongjie Liu, J. Lv, Qian Hu
The communications-based train control (CBTC) is a typical safety-critical system that protects and directs train operations in urban rail transit. It is suggested to provide on-going safety protections for the automatic train protection, which is a kernel function of the CBTC. Runtime verification is a technique for monitoring system executions against safety requirements. A particular challenge in implementing of a runtime verification system for the CBTC is the appropriate monitor specification. This paper presents a novel dynamic monitoring generation method to the problem. The train control procedures of the CBTC is formalized by parametric hybrid automata (PHA), which introduces notations of parametric expressions for flow, transition conditions and invariants. With an observation, the PHA is instantiated to a standard hybrid automaton. The monitor specification is then generated automatically by calculating the reachable set of the automaton with respect to some selected safety-related properties. The presented method is evaluated in a hard-ware in the loop CBTC platform, which is developed with realistic engineering data of Beijing Yizhuang metro line. The experiment results show that the approach is feasible, and various dangerous of the CBTC system are prevented from developing into accidents of train collisions.
{"title":"Runtime Verification of Communications-based Train Control with Parametric Hybrid Automata","authors":"Ming Chai, Haifeng Wang, Hongjie Liu, J. Lv, Qian Hu","doi":"10.1109/ITSC.2019.8917282","DOIUrl":"https://doi.org/10.1109/ITSC.2019.8917282","url":null,"abstract":"The communications-based train control (CBTC) is a typical safety-critical system that protects and directs train operations in urban rail transit. It is suggested to provide on-going safety protections for the automatic train protection, which is a kernel function of the CBTC. Runtime verification is a technique for monitoring system executions against safety requirements. A particular challenge in implementing of a runtime verification system for the CBTC is the appropriate monitor specification. This paper presents a novel dynamic monitoring generation method to the problem. The train control procedures of the CBTC is formalized by parametric hybrid automata (PHA), which introduces notations of parametric expressions for flow, transition conditions and invariants. With an observation, the PHA is instantiated to a standard hybrid automaton. The monitor specification is then generated automatically by calculating the reachable set of the automaton with respect to some selected safety-related properties. The presented method is evaluated in a hard-ware in the loop CBTC platform, which is developed with realistic engineering data of Beijing Yizhuang metro line. The experiment results show that the approach is feasible, and various dangerous of the CBTC system are prevented from developing into accidents of train collisions.","PeriodicalId":6717,"journal":{"name":"2019 IEEE Intelligent Transportation Systems Conference (ITSC)","volume":"101 ","pages":"2160-2165"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"91465208","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-10-01DOI: 10.1109/ITSC.2019.8916954
P. Sieberg, S. Blume, N. Harnack, Niko Maas, D. Schramm
This article presents a hybrid state estimation using vehicle dynamics as an application. The knowledge about the dynamic states are essential in the vehicle. Ultimately, the built-in control algorithms are using these states to exploit safety, comfort, and performance. In most cases, the states of the vehicle are measured directly. Nevertheless, direct measurement is not profitable or difficult to implement for all states of vehicle dynamics. In this case, state estimators are used. In the past, classical approaches such as modelling of the physical systems have been used for estimation. Due to the continuous developments in the field of computing hardware, methods of machine learning can now also be used in this context. The presented article includes artificial neural networks. With this method, a transfer behavior can be mapped without having knowledge about the system to be estimated. A major problem of such artificial neural networks, however, is the traceability as well as checking the robustness for universal use. Therefore, the artificial neural network is coupled with physical knowledge. This results in a hybrid state estimator based on a Kalman filter. This novel hybrid approach is presented using the example of estimating the roll angle of a vehicle.
{"title":"Hybrid State Estimation Combining Artificial Neural Network and Physical Model","authors":"P. Sieberg, S. Blume, N. Harnack, Niko Maas, D. Schramm","doi":"10.1109/ITSC.2019.8916954","DOIUrl":"https://doi.org/10.1109/ITSC.2019.8916954","url":null,"abstract":"This article presents a hybrid state estimation using vehicle dynamics as an application. The knowledge about the dynamic states are essential in the vehicle. Ultimately, the built-in control algorithms are using these states to exploit safety, comfort, and performance. In most cases, the states of the vehicle are measured directly. Nevertheless, direct measurement is not profitable or difficult to implement for all states of vehicle dynamics. In this case, state estimators are used. In the past, classical approaches such as modelling of the physical systems have been used for estimation. Due to the continuous developments in the field of computing hardware, methods of machine learning can now also be used in this context. The presented article includes artificial neural networks. With this method, a transfer behavior can be mapped without having knowledge about the system to be estimated. A major problem of such artificial neural networks, however, is the traceability as well as checking the robustness for universal use. Therefore, the artificial neural network is coupled with physical knowledge. This results in a hybrid state estimator based on a Kalman filter. This novel hybrid approach is presented using the example of estimating the roll angle of a vehicle.","PeriodicalId":6717,"journal":{"name":"2019 IEEE Intelligent Transportation Systems Conference (ITSC)","volume":"94 1","pages":"894-899"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"91328204","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-10-01DOI: 10.1109/ITSC.2019.8917117
M. Kloock, Patrick Scheffe, S. Marquardt, Janis Maczijewski, Bassam Alrifaee, S. Kowalewski
This paper investigates intersection control of multiple vehicles using a Model Predictive Control (MPC) framework. Vehicles follow pre-defined paths across the intersection and adjust their velocities to ensure collision-free passage while maximizing an objective. We choose a non-cooperative Distributed Model Predictive Control (DMPC) approach, where priorities need to be assigned to vehicles. The algorithm we present sets these priorities automatically by evaluating the vehicles’ time to react to stop before entering the intersection. We demonstrate our method in simulations of multiple vehicles and continuous traffic. It produces near-optimal velocity profiles and reduces the computation time in comparison to centralized MPC while avoiding vehicle collisions and deadlocks.
{"title":"Distributed Model Predictive Intersection Control of Multiple Vehicles","authors":"M. Kloock, Patrick Scheffe, S. Marquardt, Janis Maczijewski, Bassam Alrifaee, S. Kowalewski","doi":"10.1109/ITSC.2019.8917117","DOIUrl":"https://doi.org/10.1109/ITSC.2019.8917117","url":null,"abstract":"This paper investigates intersection control of multiple vehicles using a Model Predictive Control (MPC) framework. Vehicles follow pre-defined paths across the intersection and adjust their velocities to ensure collision-free passage while maximizing an objective. We choose a non-cooperative Distributed Model Predictive Control (DMPC) approach, where priorities need to be assigned to vehicles. The algorithm we present sets these priorities automatically by evaluating the vehicles’ time to react to stop before entering the intersection. We demonstrate our method in simulations of multiple vehicles and continuous traffic. It produces near-optimal velocity profiles and reduces the computation time in comparison to centralized MPC while avoiding vehicle collisions and deadlocks.","PeriodicalId":6717,"journal":{"name":"2019 IEEE Intelligent Transportation Systems Conference (ITSC)","volume":"7 1","pages":"1735-1740"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"87110945","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-10-01DOI: 10.1109/ITSC.2019.8917158
Zhensong Wei, Chao Wang, Peng Hao, M. Barth
Accurate lane localization and lane change detection are crucial in advanced driver assistance systems and autonomous driving systems for safer and more efficient trajectory planning. Conventional localization devices such as Global Positioning System only provide road-level resolution for car navigation, which is incompetent to assist in lane-level decision making. The state of art technique for lane localization is to use Light Detection and Ranging sensors to correct the global localization error and achieve centimeter-level accuracy, but the real-time implementation and popularization for LiDAR is still limited by its computational burden and current cost. As a cost-effective alternative, vision-based lane change detection has been highly regarded for affordable autonomous vehicles to support lane-level localization. A deep learning based computer vision system is developed to detect the lane change behavior using the images captured by a front-view camera mounted on the vehicle and data from the inertial measurement unit for highway driving. Testing results on real-world driving data have shown that the proposed method is robust with real-time working ability and could achieve around 87% lane change detection accuracy. Compared to the average human reaction to visual stimuli, the proposed computer vision system works 9 times faster, which makes it capable of helping make life-saving decisions in time.
{"title":"Vision-Based Lane-Changing Behavior Detection Using Deep Residual Neural Network","authors":"Zhensong Wei, Chao Wang, Peng Hao, M. Barth","doi":"10.1109/ITSC.2019.8917158","DOIUrl":"https://doi.org/10.1109/ITSC.2019.8917158","url":null,"abstract":"Accurate lane localization and lane change detection are crucial in advanced driver assistance systems and autonomous driving systems for safer and more efficient trajectory planning. Conventional localization devices such as Global Positioning System only provide road-level resolution for car navigation, which is incompetent to assist in lane-level decision making. The state of art technique for lane localization is to use Light Detection and Ranging sensors to correct the global localization error and achieve centimeter-level accuracy, but the real-time implementation and popularization for LiDAR is still limited by its computational burden and current cost. As a cost-effective alternative, vision-based lane change detection has been highly regarded for affordable autonomous vehicles to support lane-level localization. A deep learning based computer vision system is developed to detect the lane change behavior using the images captured by a front-view camera mounted on the vehicle and data from the inertial measurement unit for highway driving. Testing results on real-world driving data have shown that the proposed method is robust with real-time working ability and could achieve around 87% lane change detection accuracy. Compared to the average human reaction to visual stimuli, the proposed computer vision system works 9 times faster, which makes it capable of helping make life-saving decisions in time.","PeriodicalId":6717,"journal":{"name":"2019 IEEE Intelligent Transportation Systems Conference (ITSC)","volume":"28 1","pages":"3108-3113"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"87169878","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-10-01DOI: 10.1109/ITSC.2019.8917390
Victor Vaquero, Kai Fischer, F. Moreno-Noguer, A. Sanfeliu, Stefan Milz
Localization and Mapping is an essential component to enable Autonomous Vehicles navigation, and requires an accuracy exceeding that of commercial GPS-based systems. Current odometry and mapping algorithms are able to provide this accurate information. However, the lack of robustness of these algorithms against dynamic obstacles and environmental changes, even for short time periods, forces the generation of new maps on every session without taking advantage of previously obtained ones. In this paper we propose the use of a deep learning architecture to segment movable objects from 3D LiDAR point clouds in order to obtain longer-lasting 3D maps. This will in turn allow for better, faster and more accurate re-localization and trajectoy estimation on subsequent days. We show the effectiveness of our approach in a very dynamic and cluttered scenario, a supermarket parking lot. For that, we record several sequences on different days and compare localization errors with and without our movable objects segmentation method. Results show that we are able to accurately re-locate over a filtered map, consistently reducing trajectory errors between an average of 35.1% with respect to a non-filtered map version and of 47.9% with respect to a standalone map created on the current session.
{"title":"Improving Map Re-localization with Deep ‘Movable’ Objects Segmentation on 3D LiDAR Point Clouds","authors":"Victor Vaquero, Kai Fischer, F. Moreno-Noguer, A. Sanfeliu, Stefan Milz","doi":"10.1109/ITSC.2019.8917390","DOIUrl":"https://doi.org/10.1109/ITSC.2019.8917390","url":null,"abstract":"Localization and Mapping is an essential component to enable Autonomous Vehicles navigation, and requires an accuracy exceeding that of commercial GPS-based systems. Current odometry and mapping algorithms are able to provide this accurate information. However, the lack of robustness of these algorithms against dynamic obstacles and environmental changes, even for short time periods, forces the generation of new maps on every session without taking advantage of previously obtained ones. In this paper we propose the use of a deep learning architecture to segment movable objects from 3D LiDAR point clouds in order to obtain longer-lasting 3D maps. This will in turn allow for better, faster and more accurate re-localization and trajectoy estimation on subsequent days. We show the effectiveness of our approach in a very dynamic and cluttered scenario, a supermarket parking lot. For that, we record several sequences on different days and compare localization errors with and without our movable objects segmentation method. Results show that we are able to accurately re-locate over a filtered map, consistently reducing trajectory errors between an average of 35.1% with respect to a non-filtered map version and of 47.9% with respect to a standalone map created on the current session.","PeriodicalId":6717,"journal":{"name":"2019 IEEE Intelligent Transportation Systems Conference (ITSC)","volume":"8 1","pages":"942-949"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"87279672","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2019-10-01DOI: 10.1109/ITSC.2019.8917308
Ruoyu Sun, Donghao Xu, Huijing Zhao, M. Moze, F. Aioun, F. Guillemard
Learning human’s car-following behavior needs not only well-designed models but also effective training or calibration methods. Comparing with the vast amount of efforts on car-following modeling in literature, training methods are less studied. This research proposes a training method (BPTS - Backpropagation through Simulation) to reduce the long-term error of neural network-based car-following models, with multiple experimental validations. The training method uses a recurrent framework with simulation to generate long-term predictions for generic car-following models, and use gradient backpropagation to reduce accumulative error. The proposed training method can also calibrate other car-following models besides neural network-based models. In experimental validation, our studies yielded more than 30% error reduction in long-term (20 s) prediction for feed-forward Artificial Neural Network (ANN) and Long short-term memory (LSTM) models, and reduces the error on vehicle position by more than 1.0 meters, at the cost of that short-term (0.2 s) prediction error slightly increases. The proposed training method dramatically reduces the long-term prediction error of neural network-based car-following models.
{"title":"Backpropagation through Simulation: A Training Method for Neural Network-based Car-following","authors":"Ruoyu Sun, Donghao Xu, Huijing Zhao, M. Moze, F. Aioun, F. Guillemard","doi":"10.1109/ITSC.2019.8917308","DOIUrl":"https://doi.org/10.1109/ITSC.2019.8917308","url":null,"abstract":"Learning human’s car-following behavior needs not only well-designed models but also effective training or calibration methods. Comparing with the vast amount of efforts on car-following modeling in literature, training methods are less studied. This research proposes a training method (BPTS - Backpropagation through Simulation) to reduce the long-term error of neural network-based car-following models, with multiple experimental validations. The training method uses a recurrent framework with simulation to generate long-term predictions for generic car-following models, and use gradient backpropagation to reduce accumulative error. The proposed training method can also calibrate other car-following models besides neural network-based models. In experimental validation, our studies yielded more than 30% error reduction in long-term (20 s) prediction for feed-forward Artificial Neural Network (ANN) and Long short-term memory (LSTM) models, and reduces the error on vehicle position by more than 1.0 meters, at the cost of that short-term (0.2 s) prediction error slightly increases. The proposed training method dramatically reduces the long-term prediction error of neural network-based car-following models.","PeriodicalId":6717,"journal":{"name":"2019 IEEE Intelligent Transportation Systems Conference (ITSC)","volume":"1 1","pages":"3796-3803"},"PeriodicalIF":0.0,"publicationDate":"2019-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"87295975","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","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：