Mobil Robotlarda Parçacık Filtresi Kullanarak Eş Zamanlı Lokalizasyon Ve Haritalama
Yükleniyor...
Dosyalar
Tarih
item.page.authors
Süreli Yayın başlığı
Süreli Yayın ISSN
Cilt Başlığı
Yayınevi
Fen Bilimleri Enstitüsü
Institute of Science and Technology
Institute of Science and Technology
Özet
Parçacık filtresi, genişletilmiş Kalman filtresi(GKF) gibi doğrusal olmayan denklemleri, doğrusal gauss tipi denklemlere yakınsatarak SLAM problemini çözmek yerine, gauss tipi olmayan süreçleri ve dağılımları içeren ardışıl Monte Carlo tekniklerini kullanır. Tezin amacı, parçacık filtresinin SLAM problemindeki başarısını ölçmektir. Tezde popüler olarak tercih edilen parçacık filtrelerinin özelleştirilmiş bir modeli olan, Rao-Blackwellisation adlı variyans azaltma tekniğini içinde barındıran FastSLAM adlı Rao-Blackwellised parçacık filtreleri incelenmiştir. Her parçacık, çoklu veri eşleştirmesine izin veren bir biçimde, çevre haritasını ve robot lokalizasyonunu saklar. Mobil robotun yer bulma probleminde Monte Carlo lokalizasyonu(MCL) tarafından işaretçi nesnelerin konumlarının kestirimi ise GKF ile yapılır. Bu yapı ile gauss tipi işaretçi nesne kestirimi parçacık filtresine kazandırılmış olur. RBPF’nin anlaşılması açısından iteratif Bayes kestirimi, Kalman filtresi, MCL, maksimum benzerlik kestirimi konuları açıklanmış ve filtrenin adımlarını olan örnekleme, ardışıl önem ağırlıkları hesaplaması, haritalama, yeniden örnekleme ve kriterlerinden bahsedilmiştir.
Particle filters which contain non-linear Gaussian process and distributions use sequential Monte Carlo techniques instead of approximating the non-linear, non-Gaussian process to the linear Gaussian ones such as made by the extended Kalman filter (EKF). The purpose of the thesis is to measure the success rate of the particle filters for the SLAM problem. In the thesis, popularly used and specialized form of the particle filters, the FastSLAM algorithm in which a variance reduction technique named as Rao-Blackwellisation is found, is observed. FastSLAM divides the SLAM problem into two categories such as the robotic mapping problem and the estimation of the locations of the landmarks. Each particle of the algorithm stores the landmarks’ map and the robot pose rendering the multiple hypothesis data associations. The robot pose estimation is computed by Monte Carlo localization (MCL) algorithms and the coordinates and their uncertainties of the landmarks are estimated by the EKF. This model implements the Gaussian estimation of the observations into the particle filters. Sequential Bayes estimation, Kalman filter, MCL, maximum likelihood estimation and the application steps which are sampling, importance weighting, mapping, resampling and its criteria are included in the thesis for a better comprehension of RBPF.
Particle filters which contain non-linear Gaussian process and distributions use sequential Monte Carlo techniques instead of approximating the non-linear, non-Gaussian process to the linear Gaussian ones such as made by the extended Kalman filter (EKF). The purpose of the thesis is to measure the success rate of the particle filters for the SLAM problem. In the thesis, popularly used and specialized form of the particle filters, the FastSLAM algorithm in which a variance reduction technique named as Rao-Blackwellisation is found, is observed. FastSLAM divides the SLAM problem into two categories such as the robotic mapping problem and the estimation of the locations of the landmarks. Each particle of the algorithm stores the landmarks’ map and the robot pose rendering the multiple hypothesis data associations. The robot pose estimation is computed by Monte Carlo localization (MCL) algorithms and the coordinates and their uncertainties of the landmarks are estimated by the EKF. This model implements the Gaussian estimation of the observations into the particle filters. Sequential Bayes estimation, Kalman filter, MCL, maximum likelihood estimation and the application steps which are sampling, importance weighting, mapping, resampling and its criteria are included in the thesis for a better comprehension of RBPF.
Açıklama
Tez (Yüksek Lisans) -- İstanbul Teknik Üniversitesi, Fen Bilimleri Enstitüsü, 2009
Thesis (M.Sc.) -- İstanbul Technical University, Institute of Science and Technology, 2009
Thesis (M.Sc.) -- İstanbul Technical University, Institute of Science and Technology, 2009
Konusu
Mobil robot, filtre, lokalizasyon, haritalama, SLAM, FastSLAM, Bayes, Monte Carlo, Rao-Blacwellised, Mobile robot, filter, localization, mapping, SLAM, FastSLAM, Bayes, Monte Carlo, Rao-Blacwellised
