東華大學圖書館 |

Data-Efficient Methods for Model Learning and Control in Robotics = = Efektivni Metody pro Uceni Modelu a Rizeni v Robotice.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Data-Efficient Methods for Model Learning and Control in Robotics =/
其他題名:	Efektivni Metody pro Uceni Modelu a Rizeni v Robotice.
其他題名:	Efektivni Metody pro Uceni Modelu a Rizeni v Robotice.
作者:	Derner, Erik.
面頁冊數:	1 online resource (126 pages)
附註:	Source: Dissertations Abstracts International, Volume: 83-12, Section: B.
Contained By:	Dissertations Abstracts International83-12B.
標題:	Robots. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=29117982click for full text (PQDT)
ISBN:	9798802729335

Data-Efficient Methods for Model Learning and Control in Robotics = = Efektivni Metody pro Uceni Modelu a Rizeni v Robotice.
Derner, Erik.

Data-Efficient Methods for Model Learning and Control in Robotics =Efektivni Metody pro Uceni Modelu a Rizeni v Robotice.Efektivni Metody pro Uceni Modelu a Rizeni v Robotice. - 1 online resource (126 pages)

Source: Dissertations Abstracts International, Volume: 83-12, Section: B.

Thesis (Ph.D.)--Czech Technical University, 2022.

Includes bibliographical references

Constructing mathematical models of dynamic systems is central to many engineering and science disciplines. Models facilitate simulations, analysis of the system's behavior, decision making, and design of automatic control algorithms. Even inherently model-free control techniques such as reinforcement learning have been shown to benefit from the use of models. However, applying model learning methods to robotics is not straightforward. Obtaining informative data for constructing dynamic models can be difficult, especially when the models are to be learned during task execution. Despite their increasing popularity, commonly used model learning methods such as deep neural networks come with drawbacks. They are datahungry and require a lot of computational power to learn a large number of parameters in their complex structure. Their black-box nature does not offer any insight into or interpretation of the model. Also, configuring these methods to achieve good results is often a difficult task.The objective of this thesis is to address the present challenges in data-driven model learning in robotics. Several variants and extensions of symbolic regression are introduced. This technique, based on genetic programming, is suitable to automatically build compact and accurate models in the form of analytic equations even from small data sets. One of the challenges is posed by the large amount of data the robots collect during their operation, demanding techniques to select a smaller subset of training samples. To that end, this thesis presents a novel sample-selection method based on model prediction error and compares it to four alternative approaches. A real-world experimental evaluation on a mobile robot shows that a model learned from only a few tens of samples selected by the proposed method can be used to accomplish a motion control task within a reinforcement learning scheme.Standard data-driven model learning techniques in many cases yield models that violate the physical constraints of the robot. However, a partial theoretical or empirical model of the robot is often known. It is shown in this work how symbolic regression can be naturally extended to include the prior information into the model construction process. An experimental evaluation on two real-world robotic platforms demonstrates that symbolic regression is able to automatically build models that are both accurate and physically valid and compensate for theoretical or empirical model deficiencies.Efficient methods are needed not only to learn robot models but also to learn models of the robot's environment. The thesis is concluded by presenting a novel method for reliable robot localization in dynamic environments. The proposed approach introduces an environment representation based on weighted local visual features and a change detection algorithm that updates the weights as the robot moves around the environment. The core idea of the method consists in using the weights to distinguish the useful information in stable regions of the scene from the unreliable information in the regions that are changing. An extensive evaluation and comparison to state-of-the-art alternatives show that using the proposed change detection algorithm improves the localization accuracy.

Electronic reproduction.
Ann Arbor, Mich. :
ProQuest,
2023

Mode of access: World Wide Web

ISBN: 9798802729335Subjects--Topical Terms:

529507
Robots.
Index Terms--Genre/Form:

542853
Electronic books.

Data-Efficient Methods for Model Learning and Control in Robotics = = Efektivni Metody pro Uceni Modelu a Rizeni v Robotice.
LDR:07771nmm a2200397K 4500 001 2354167
005 20230324111216.5
006 m o d
007 cr mn ---uuuuu
008 241011s2022 xx obm 000 0 eng d
020 $a 9798802729335
035 $a (MiAaPQ)AAI29117982
035 $a (MiAaPQ)CzechTech99977
035 $a AAI29117982
040 $a MiAaPQ $b eng $c MiAaPQ $d NTU
100 1 $a Derner, Erik. $3 3694515
245 1 0 $a Data-Efficient Methods for Model Learning and Control in Robotics = $b Efektivni Metody pro Uceni Modelu a Rizeni v Robotice.
246 3 1 $a Efektivni Metody pro Uceni Modelu a Rizeni v Robotice.
264 0 $c 2022
300 $a 1 online resource (126 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 83-12, Section: B.
500 $a Advisor: Babuska, Robert.
502 $a Thesis (Ph.D.)--Czech Technical University, 2022.
504 $a Includes bibliographical references
520 $a Constructing mathematical models of dynamic systems is central to many engineering and science disciplines. Models facilitate simulations, analysis of the system's behavior, decision making, and design of automatic control algorithms. Even inherently model-free control techniques such as reinforcement learning have been shown to benefit from the use of models. However, applying model learning methods to robotics is not straightforward. Obtaining informative data for constructing dynamic models can be difficult, especially when the models are to be learned during task execution. Despite their increasing popularity, commonly used model learning methods such as deep neural networks come with drawbacks. They are datahungry and require a lot of computational power to learn a large number of parameters in their complex structure. Their black-box nature does not offer any insight into or interpretation of the model. Also, configuring these methods to achieve good results is often a difficult task.The objective of this thesis is to address the present challenges in data-driven model learning in robotics. Several variants and extensions of symbolic regression are introduced. This technique, based on genetic programming, is suitable to automatically build compact and accurate models in the form of analytic equations even from small data sets. One of the challenges is posed by the large amount of data the robots collect during their operation, demanding techniques to select a smaller subset of training samples. To that end, this thesis presents a novel sample-selection method based on model prediction error and compares it to four alternative approaches. A real-world experimental evaluation on a mobile robot shows that a model learned from only a few tens of samples selected by the proposed method can be used to accomplish a motion control task within a reinforcement learning scheme.Standard data-driven model learning techniques in many cases yield models that violate the physical constraints of the robot. However, a partial theoretical or empirical model of the robot is often known. It is shown in this work how symbolic regression can be naturally extended to include the prior information into the model construction process. An experimental evaluation on two real-world robotic platforms demonstrates that symbolic regression is able to automatically build models that are both accurate and physically valid and compensate for theoretical or empirical model deficiencies.Efficient methods are needed not only to learn robot models but also to learn models of the robot's environment. The thesis is concluded by presenting a novel method for reliable robot localization in dynamic environments. The proposed approach introduces an environment representation based on weighted local visual features and a change detection algorithm that updates the weights as the robot moves around the environment. The core idea of the method consists in using the weights to distinguish the useful information in stable regions of the scene from the unreliable information in the regions that are changing. An extensive evaluation and comparison to state-of-the-art alternatives show that using the proposed change detection algorithm improves the localization accuracy.
520 $a Znalost matematickych modelu dynamickych systemu je klicova pro celou radu inzenyrskych a vedeckych disciplin. Modely umoznuji provadeni simulaci, analyzu chovani systemu, rozho- dovani a navrh ridicich algoritmu. Z pouziti modelu tezi i techniky, ktere z principu funguji bez modelu, napriklad posilovane uceni. Vyuziti metod pro uceni modelu v robotice ma vsak sva specifika. Ziskat informativni data pro uceni dynamickych modelu muze byt obtizne, zvlaste behem vykonavani dane ulohy. Navzdory rostouci popularite maji bezne pouzivane metody uceni modelu, jako jsou hluboke neuronove site, sve nevyhody. Vyzaduji velky objem treno- vacich dat a znacny vypocetni vykon, aby se naucily velky pocet parametru. Jejich black-box charakter neumoznuje interpretaci modelu ani vhled do jeho struktury. Take spravne nasta- veni konfigurace pro dosazeni dobrych vysledku je u techto metod casto obtizny ukol.Cilem teto disertacni prace je navrhnout reseni aktualnich problemu v oblasti uceni mo- delu z dat v robotice. Prace predstavuje nekolik variant a rozsireni symbolicke regrese. Tato technika, zalozena na genetickem programovani, je vhodna pro automaticke vytvareni kom- paktnich a presnych modelu v podobe analytickych rovnic i z malych souboru dat. Jednim z problemu v robotice je velke mnozstvi dat, ktere jsou roboty behem provozu shromazdovany, coz vyzaduje vyber podmnoziny trenovacich vzorku. Tato prace predstavuje novou metodu vyberu vzorku zalozenou na predikcni chybe modelu a porovnava ji se ctyrmi alternativnimi metodami. Experimentalni vyhodnoceni na mobilnim robotu ukazuje, ze model nauceny jen z nekolika desitek vzorku vybranych navrzenou metodou muze byt vyuzit pro uspesne vyko- nani ulohy zalozene na rizeni metodou posilovaneho uceni.Bezne pouzivane techniky uceni modelu z dat v mnoha pripadech generuji modely, ktere nevyhovuji fyzikalnim omezenim robota. Castecny teoreticky nebo empiricky model robota je pritom casto znam. Tato prace ukazuje, jak lze symbolickou regresi prirozene rozsirit tak, aby byly predem zname informace o robotu zahrnuty do procesu uceni modelu. Experimen- talni vyhodnoceni na dvou ruznych robotech ukazuje, ze symbolicka regrese je schopna au- tomaticky vytvaret modely, ktere jsou presne, vyhovuji fyzikalnim omezenim a kompenzuji nedostatky teoretickeho nebo empirickeho modelu.Efektivni metody jsou treba nejen k uceni modelu robotu, ale take k uceni modelu prostredi robota. Prace je zakoncena predstavenim nove metody pro spolehlivou lokalizaci robotu v dy- namickych prostredich. V navrhovanem pristupu se vyuziva model prostredi zalozeny na va- zenych lokalnich vizualnich priznacich. Algoritmus detekce zmen prubezne aktualizuje tyto vahy behem pohybu robota prostredim. Zakladni myslenkou metody je na zaklade techto vah rozlisit uzitecne informace ve stabilnich oblastech sceny od nespolehlivych informaci v oblas- tech, ktere se meni. Rozsahle experimentalni vyhodnoceni a srovnani s alternativnimi meto- dami ukazuje, ze pouziti navrzeneho algoritmu detekce zmen zlepsuje presnost lokalizace.
533 $a Electronic reproduction. $b Ann Arbor, Mich. : $c ProQuest, $d 2023
538 $a Mode of access: World Wide Web
650 4 $a Robots. $3 529507
650 4 $a Control algorithms. $3 3560702
650 4 $a Collaboration. $3 3556296
650 4 $a Teaching methods. $3 595505
650 4 $a Localization. $3 3560711
650 4 $a Neural networks. $3 677449
650 4 $a Robotics. $3 519753
650 4 $a Artificial intelligence. $3 516317
650 4 $a Computer science. $3 523869
650 4 $a Education. $3 516579
650 4 $a Pedagogy. $3 2122828
655 7 $a Electronic books. $2 lcsh $3 542853
690 $a 0771
690 $a 0800
690 $a 0984
690 $a 0515
690 $a 0456
710 2 $a ProQuest Information and Learning Co. $3 783688
710 2 $a Czech Technical University. $3 3694516
773 0 $t Dissertations Abstracts International $g 83-12B.
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=29117982 $z click for full text (PQDT)