Search

To be able to navigate autonomously, a vehicle must be accurately localized relatively to all obstacles, such as roadside for lane keeping and vehicles and pedestrians to avoid causing accidents. This PhD thesis deals with the interest of cooperation to improve the localization of cooperative vehicles that exchange information. Autonomous navigation on the road is often based on coordinates provided in a Cartesian frame. In order to better represent the pose of a vehicle with respect to the lane in which it travels, we study curvilinear coordinates with respect to a path stored in a map. These coordinates generalize the curvilinear abscissa by adding a signed lateral deviation from the center of the lane and an orientation relative to the center of the lane taking into account the direction of travel. These coordinates are studied with different track models and using different projections to make the map-matching. A first cooperative localization approach is based on these coordinates. The lateral deviation and the orientation relative to the lane can be known precisely from a perception of the lane borders, but for autonomous driving with other vehicles, it is important to maintain a good longitudinal accuracy. A one-dimensional data fusion method makes it possible to show the interest of the cooperative localization in this simplified case where the lateral deviation, the curvilinear orientation and the relative positioning between two vehicles are accurately known. This case study shows that, in some cases, lateral accuracy can be propagated to other vehicles to improve their longitudinal accuracy. The correlation issues of the errors are taken into account with a covariance intersection filter. An ICP (Iterative Closest Point) minimization algorithm is then used to determine the relative pose between the vehicles from LiDAR points and a 2D polygonal model representing the shape of the vehicle. Several correspondences of the LiDAR points with the model and different minimization approaches are compared. The propagation of absolute vehicle pose using relative poses with their uncertainties is done through non-linear equations that can have a strong impact on consistency. The different dynamic elements surrounding the ego-vehicle are estimated in a Local Dynamic Map (LDM) to enhance the static high definition map describing the center of the lane and its border. In our case, the agents are only communicating vehicles. The LDM is composed of the state of each vehicle. The states are merged using an asynchronous algorithm, fusing available data at variable times. The algorithm is decentralized, each vehicle computing its own LDM and sharing it. As the position errors of the GNSS receivers are biased, a marking detection is introduced to obtain the lateral deviation from the center of the lane in order to estimate these biases. LiDAR observations with the ICP method allow to enrich the fusion with the constraints between the vehicles. Experimental results of this fusion show that the vehicles are more accurately localized with respect to each other while maintaining consistent poses.; Afin de naviguer en autonomie un véhicule doit être capable de se localiser précisément par rapport aux bords de voie pour ne pas sortir de celle-ci et par rapport aux véhicules et piétons pour ne pas causer d'accident. Cette thèse traite de l'intérêt de la communication dans l'amélioration de la localisation des véhicules autonomes. La navigation autonome sur route est souvent réalisée à partir de coordonnées cartésiennes. Afin de mieux représenter la pose d'un véhicule relativement à la voie dans laquelle il circule, nous étudions l'utilisation de coordonnées curvilignes le long de chemins enregistrés dans des cartes. Ces coordonnées généralisent l'abscisse curviligne en y ajoutant un écart latéral signé par rapport au centre de la voie et une orientation relative au centre de cette voie en prenant en compte le sens de circulation. Une première approche de localisation coopérative est réalisée à partir de ces coordonnées. Une fusion de données à une dimension permet de montrer l'intérêt de la localisation coopérative dans le cas simplifié où l'écart latéral, l'orientation curviligne et la pose relative entre deux véhicules sont connus avec précision. Les problèmes de corrélation des erreurs dus à l'échange d'information sont pris en compte grâce à un filtre par intersection de covariance. Nous présentons ensuite à une méthode de perception de type ICP (Iterative Closest Point) pour déterminer la pose relative entre les véhicules à partir de points LiDAR et d'un modèle polygonal 2D représentant la forme du véhicule. La propagation des erreurs de poses absolues des véhicules à l'aide de poses relatives estimées avec leurs incertitudes se fait via des équations non linéaires qui peuvent avoir un fort impact sur la consistance. Les poses des différents véhicules entourant l'égo-véhicule sont estimés dans une carte locale dynamique (CLD) permettant d'enrichir la carte statique haute définition décrivant le centre de la voie et les bords de celle-ci. La carte locale dynamique est composée de l'état de chaque véhicule communicant. Les états sont fusionnés en utilisant un algorithme asynchrone, à partir de données disponibles à des temps variables. L'algorithme est décentralisé, chaque véhicule calculant sa propre CLD et la partageant. Les erreurs de position des récepteurs GNSS étant biaisées, une détection de marquages est introduite pour obtenir la distance latérale par rapport au centre de la voie afin d'estimer ces biais. Des observations LiDAR avec la méthode ICP permettent de plus d'enrichir la fusion avec des contraintes entre les véhicules. Des résultats expérimentaux illustrent les performances de cette approche en termes de précision et de consistance.

To be able to navigate autonomously, a vehicle must be accurately localized relatively to all obstacles, such as roadside for lane keeping and vehicles and pedestrians to avoid causing accidents. This PhD thesis deals with the interest of cooperation to improve the localization of cooperative vehicles that exchange information. Autonomous navigation on the road is often based on coordinates provided in a Cartesian frame. In order to better represent the pose of a vehicle with respect to the lane in which it travels, we study curvilinear coordinates with respect to a path stored in a map. These coordinates generalize the curvilinear abscissa by adding a signed lateral deviation from the center of the lane and an orientation relative to the center of the lane taking into account the direction of travel. These coordinates are studied with different track models and using different projections to make the map-matching. A first cooperative localization approach is based on these coordinates. The lateral deviation and the orientation relative to the lane can be known precisely from a perception of the lane borders, but for autonomous driving with other vehicles, it is important to maintain a good longitudinal accuracy. A one-dimensional data fusion method makes it possible to show the interest of the cooperative localization in this simplified case where the lateral deviation, the curvilinear orientation and the relative positioning between two vehicles are accurately known. This case study shows that, in some cases, lateral accuracy can be propagated to other vehicles to improve their longitudinal accuracy. The correlation issues of the errors are taken into account with a covariance intersection filter. An ICP (Iterative Closest Point) minimization algorithm is then used to determine the relative pose between the vehicles from LiDAR points and a 2D polygonal model representing the shape of the vehicle. Several correspondences of the LiDAR points with the model and different minimization approaches are compared. The propagation of absolute vehicle pose using relative poses with their uncertainties is done through non-linear equations that can have a strong impact on consistency. The different dynamic elements surrounding the ego-vehicle are estimated in a Local Dynamic Map (LDM) to enhance the static high definition map describing the center of the lane and its border. In our case, the agents are only communicating vehicles. The LDM is composed of the state of each vehicle. The states are merged using an asynchronous algorithm, fusing available data at variable times. The algorithm is decentralized, each vehicle computing its own LDM and sharing it. As the position errors of the GNSS receivers are biased, a marking detection is introduced to obtain the lateral deviation from the center of the lane in order to estimate these biases. LiDAR observations with the ICP method allow to enrich the fusion with the constraints between the vehicles. Experimental results of this fusion show that the vehicles are more accurately localized with respect to each other while maintaining consistent poses.; Afin de naviguer en autonomie un véhicule doit être capable de se localiser précisément par rapport aux bords de voie pour ne pas sortir de celle-ci et par rapport aux véhicules et piétons pour ne pas causer d'accident. Cette thèse traite de l'intérêt de la communication dans l'amélioration de la localisation des véhicules autonomes. La navigation autonome sur route est souvent réalisée à partir de coordonnées cartésiennes. Afin de mieux représenter la pose d'un véhicule relativement à la voie dans laquelle il circule, nous étudions l'utilisation de coordonnées curvilignes le long de chemins enregistrés dans des cartes. Ces coordonnées généralisent l'abscisse curviligne en y ajoutant un écart latéral signé par rapport au centre de la voie et une orientation relative au centre de cette voie en prenant en compte le sens de circulation. Une première approche de localisation coopérative est réalisée à partir de ces coordonnées. Une fusion de données à une dimension permet de montrer l'intérêt de la localisation coopérative dans le cas simplifié où l'écart latéral, l'orientation curviligne et la pose relative entre deux véhicules sont connus avec précision. Les problèmes de corrélation des erreurs dus à l'échange d'information sont pris en compte grâce à un filtre par intersection de covariance. Nous présentons ensuite à une méthode de perception de type ICP (Iterative Closest Point) pour déterminer la pose relative entre les véhicules à partir de points LiDAR et d'un modèle polygonal 2D représentant la forme du véhicule. La propagation des erreurs de poses absolues des véhicules à l'aide de poses relatives estimées avec leurs incertitudes se fait via des équations non linéaires qui peuvent avoir un fort impact sur la consistance. Les poses des différents véhicules entourant l'égo-véhicule sont estimés dans une carte locale dynamique (CLD) permettant d'enrichir la carte statique haute définition décrivant le centre de la voie et les bords de celle-ci. La carte locale dynamique est composée de l'état de chaque véhicule communicant. Les états sont fusionnés en utilisant un algorithme asynchrone, à partir de données disponibles à des temps variables. L'algorithme est décentralisé, chaque véhicule calculant sa propre CLD et la partageant. Les erreurs de position des récepteurs GNSS étant biaisées, une détection de marquages est introduite pour obtenir la distance latérale par rapport au centre de la voie afin d'estimer ces biais. Des observations LiDAR avec la méthode ICP permettent de plus d'enrichir la fusion avec des contraintes entre les véhicules. Des résultats expérimentaux illustrent les performances de cette approche en termes de précision et de consistance.

The work describes the contribution of SAFESTRIP EU project technology to the implementation of the Dynamic Local Map used by C-ITS safety applications. The road strips developed in SAFESTRIP are able to detect and estimate the longitudinal and lateral position of detected vehicle at lane level. This is exploited by many existing and new C-ITS applications. Here a Coorperative Intersection Support application is described and used as example to explain the concept and highlight the benefit put forward by the road strips informations.

This paper presents the rationale and design of the DynaMap, a Local Dynamic Map (LDM) for infrastructure-based cooperative applications on Road side and Central ITS Stations. Most LDMs are developed primarily for use in cooperative vehicles. The requirements for infrastructure-based applications are different from vehicle-based applications though. This paper motivates the requirements for road side applications and presents the architecture of the DynaMap.

This paper presents the rationale and design of the DynaMap, a Local Dynamic Map (LDM) for infrastructure-based cooperative applications on Road side and Central ITS Stations. Most LDMs are developed primarily for use in cooperative vehicles. The requirements for infrastructure-based applications are different from vehicle-based applications though. This paper motivates the requirements for road side applications and presents the architecture of the DynaMap.

1 Introduction||2 Literature Review||3 Methodology||4 Results||5 Discussion||6 Summary, Autonomous vehicles （AVs） have been increasing rapidly on the road in recent years. However, the safety of AVs is of significant concern, which we must ensure. AVs use sensor information to achieve autonomy, but sensors such as cameras and lidar have limitations, and vehicles cannot rely on them entirely for safe navigation. To assist AVs with static information, high-definition maps （HD maps） can facilitate the complex static details of the surrounding for safe autonomy. However, we can model complex static information using HD maps for navigation; detecting and maintaining the traffic participant’s dynamic information using sensors of the ego vehicle alone is still a significant concern for safe navigation. In such a situation of sensing limitations, Cooperative Intelligent Transport Systems （C-ITS） is one approach to facilitate vehicle navigation through sharing information between the traffic participants. The C-ITS approach has various Intelligent transportation system （ITS） station units, namely Personal, Vehicle, Road-side and Central ITS station units. A Local Dynamic Map （LDM） is a critical component in any ITS station’s facilities layer. LDM is one way to maintain static and dynamic information of the traffic participants in a consistent geometrical way. It is a necessary facility in C-ITS to share sensor information between participating traffic agents. Moreover, it maintains information about the objects that are either part of the traffic or influenced by it. The International Organization for Standardization （ISO） and European Telecommunications Standards Institute （ETSI） have also made standardization efforts. Since its inception in the SAFESPOT project, implementations of LDM have been mostly four-layer data organizations. Where Layer 1 and Layer 2 maintain static information and transient static information. Then, Layer 3 and Layer 4 contain transient dynamic and highly dynamic data. Depending upon the requirement, the LDM community realized memory-based or database-based LDM. We utilized the decision diagram to enhance the safety aspect of the traffic participants in the memory/ database-based LDM setup. We utilized Shared Binary Decision Diagram （SBDD） and Geohash granular properties to detect the near-miss situation, i.e. when vehicles come very close. However, besides DynaMap, there is also a common understanding since the SAFESPOT project introduced LDM to use the database and supported query language to retrieve data from the LDM. Hence, most implementations use different databases and query languages to execute it. Although, the LDM community has explored LDM depending on the database variants. Nevertheless, remarkably less emphasis has been given to the type of data stored in the LDM. This thesis attempted to fill this gap in the LDM to enhance the moving vehicle’s safety aspect. We proposed a novel method of data representation for vehicle future geographical occupancy information using a binary decision diagram （BDD）. We show that sharing BDD-based information is consistent with the C-ITS nature of the data sharing since the algebraic operation between the exchanged BDDs can confirm the possibility of future interaction. We calculated potential future occupancy using Kamm’s circle, shown in the ROS-based simulator and modified the mid-point circle generation algorithm to find the BDD representing the set of Geohash enclosing the Kamm’s circle. We also reported data insertion and collision avoidance check time of the linked list-based BDD on PostgreSQL database-based LDM., 九州工業大学博士学位論文 学位記番号：生工博甲第449号 学位授与年月日：令和4年9月26日, 令和4年度