Otonom Araçların Yöngüdümünde PAF Tabanlı EZKH Yönteminin Geliştirilmesi

Abstract

Günümüzde özellikle askeri uygulamalarda operatörsüz platform kullanımı düşüncesi yaygınlaşmış olmakla birlikte özellikle insansız ya da otonom araçlar geliştirilme amaçlarına uygun olarak verilen görevin tamamını veya bir bölümünü kendi kendine yapması beklenen araçlar olarak ortaya çıkmıştır. Bu isteği karşılayarak güvenli bir otonom yöngüdüm yapabilmenin ön koşulu ise insansız aracının konumunun doğru şekilde bilinmesidir. Günümüzde konum belirlemenin en bilinen yöntemi Küresel Uydu Yönbulum Sistemlerinin (KUYS) kullanımıdır. KUYS'nin olmadığı veya erişilemediği ortamlarda araç konumunun ya da yöngüdüm yapılan bölgenin haritasının da bilenmemesi durumunda güvenli konum belirlemek oldukça zor bir problemdir. Eş Zamanlı Konumlama ve Haritalama (EZKH) aynı anda hem platform pozisyonu hem de ortam haritası belirlenmesi yöntemine verilen addır. Son çeyrek yüzyılda ortaya çıkan EZKH problemi, 2000'li yıllardan başlayarak kara, deniz, hava araçları için araştırılmış olmakla birlikte Kalman Filtresi (KF) tabanlı Genişletilmiş Kalman Filtresi (DKF) ve Dağıtılmış Kalman Filtresi (GKF) gibi parametrik filtre yaklaşımları yanında Parçacık Filtresi (PF) gibi non-parametrik metotlardan oluşan durum tahmin yöntemleri ve model ya da grafik tabanlı üst seviye kontrol amaçlayan ve özellikle de görüntü işleyen teknikler bu kapsamda kullanılmıştır. İnsansız sistemlerin; platform-araç, sensör tipi ve kara, deniz, hava gibi ortam türü başlıklarında oldukça fazla farklılıklar göstermesi nedeniyle, EZKH probleminde genel bir bakış sağlamak için sınıflandırma (taxonomy) yoluyla performans analizi ihtiyacından bahsedilebilir. Bu çalışmada belirsizlik altında tahmin araçlarının performans analizleri ile birlikte literatürde ilk kez olarak Parçacık Akış Filtresi (PAF) tabanlı bir EZKH yapısı filtrenin matematik temelleri, filtre analizleri, otonom araç sistemi ile sensör modellerini de içerecek şekilde verilmiştir. Benzetim sonuçlarına göre parçacık akış filtresi tabanlı EZKH performansının hesaplama maliyeti nedeniyle bazı gerçek zamanlı uygulamalardaki zorluklarına rağmen literatürde daha önce yer almış diğer tahmin yöntemleriyle karşılaştırıldığında özellikle belirsizlikleri daha düşük algılayıcılar kullanan ölçüm ortamlarında parçacık filtresi yapısında ortaya çıkan dejenerasyon sorununu ortadan kaldırarak daha başarılı sonuçlar verdiği ve bu nedenle tercih edilebileceği görülmüştür.

Unmanned systems have quite importance for decision makers and operational agents due to their extensive usage and large variety of its applications such as reconnaissance and surveillance in the military and civilian areas. All but particularly military systems seeks some advances that cancel operator need. Researchers from all kind of engineering subjects are interested in unmanned aerial, territorial or marital vehicles which are going to be unique systems in near future and their navigation, control as well. Autonomous navigation of unmanned systems is also related with optimal and collision-free path planning, localizaiton and mapping or navigation. Unmanned vehicles needs remote control in case of communication failure or prevention, electromagnetic interference etc. While researches of early version of autonomous for example aerial vehicles just look for providing self take-off and landing, recent researchers expect modern unmanned vehicles to carry all duty such as surveillance, fight and so on. For this purpose autonomous cooperative navigation which succeed on obstacle avoidance is also subject of these research series. On the other hand, main usage purpose of unmanned systems is particularly autonomous navigation that provides flexible functionality notably on a standalone drive, fot this reason it is crucial to determine the unmanned vehicle's precise position for better navigation in the autonomous operations. The map knowledge allows vehicle determine its position by its sensors or with the precise position information unmanned system specify map (vice versa), which emphasizes that the position determination and mapping are indispensable steps for a prosperous autonomous navigation. The systems that determine position or map knowledge precisely may predict the other even if it has slight divergence but may not be able if the position is not certain since the position error may grow up which may cause get lost of vehicle either. Autonomous navigation, without any intervention except determination of the trajectory pattern, is an essential requirements of accurate position determination which can be done in different ways with or without GNSS that is widely used. In such system, the position is determined by the received data from GNSS sensor. GNSS, accepted as a way of safe navigation, is still used as a reliable positioning system since location and destination coordinates are known although some interruptions may occur due to its fundementals and some natural barriers. The location cannot be defined with GNSS in case of obstacles such as being indoors or under water as signal transmission is impossible and in some cases prevention or jamming of GNSS is possible. A position may also be able to determined through odometry system in GNSS denied media. Since it produces more error than GNSS it needs error correction with systems such as GNSS or one of that determine the position through stars and landforms by using map information, but in case of lack of map information systems, to determine the unknown location using the map information becomes impossible to be applied and the SLAM problem becomes more complicated. Global Navigation Satellite System (GNSS) which is actually preferred geolocation tools in the world is also mostly adequate for position determination process of autonomous vehicles such as AGV, UAV and AUV's, however it can be hard to have successful results for the same process in GNSS denied environments. Thus, there exist great deal of works in the literature about this problem recently. In GNSS denied environments (even) if the map knowledge does not exist either, the problem of simultaneous position determination and production of map information can be solved by Simultaneous Localization And Mapping (SLAM). The Simultaneous Localization and Mapping (SLAM), determining both agent's location and its surroundings at the same time, is principle for many different autonomous applications particularly in Global Navigation Satellite System denied environments. Although SLAM problem, which emerged in the last quarter of the century, has been adapted for territorial, naval and aerial platforms starting from the year of 2000's. Some parametric filter approaches such as Kalman Filter based Extended Kalman Filter and Distributed Kalman Filter, the state estimation methods including nonparametric ones such as Particle Filter, some high level control aspiring, model or graphics-based and particularly image processing techniques has been used along with it. Although there has been several attempts to find better estimation algorithms, it is still a considerable challenge to manage robust SLAM. A strong need for performance analysis of the SLAM problem by classification may be mentioned, as it varies considerably in the platform, vehicle, sensor, and media type such as territorial, naval and aerial platforms. On the other hand, non-linear filtering especially for high-dimensional systems is an important problem since failure in previous filters performances. Daum and Huang introduced an alternative non-linear filtering approach such that a homotopy which defines a particle flow introduced between the logarithms of the unnormalized prior and posterior densities at each time step of filter as the solution to a partial differential equation. The particle flow causes particles to migrate to regions where the posterior is large in value but reliance on parallel EKF or UKF exists. The particle flow filter, emerged in last decade, was particularly attractive due to its advantages such as high accuracy and fast convergence. In this research, a Particle Flow Filter based SLAM structure is given for the first time in the literature, including mathematical bases/background of the filter, analysis, an autonomous ground vehicle and a sensor model. According to the simulation results provided with the performance analysis of estimation under uncertainty tools/algorithms, although it has some computational complexity that may cause real time application concerns, the particle flow filter based SLAM performance is superior than other recursive state estimation approaches emerged before in the literature in terms of accuracy. Especially in the measurement environments with less uncertain sensors, the particle flow filter is preferable because it removes the problem of degeneration which arises in the particle filter structure.