Η παρούσα διδακτορική διατριβή έχει ως κύριο αντικείμενο μελέτης την μοντελοποίηση και έλεγχο ενός μικρορομποτικού βραχίονα αναλυόμενου σε σφαιρικά πεπερασμένα στοιχεία σε περιβάλλον με εξωτερικές δυνάμεις Van Der Waals και συνυπολογίζοντας την τριβή. Τα κύρια σημεία είναι η εισαγωγή των εξωτερικών δυνάμεων στο μοντέλο του μικρορομπότ, η δημιουργία προσαρμοστικού ελέγχου για την επίτευξη ακολουθίας τροχιάς με αναγνώριση και ακύρωση των ισχυρών μεταβαλλόμενων εξωτερικών δυνάμεων, η αναγνώριση της θέσης και η αποφυγή εμποδίων σε άγνωστο περιβάλλον κλίμακας μικρομέτρων και ο καθορισμός τροχιάς για προσέγγιση σημείων στον χώρο εργασίας του μικρορομπότ. Προτείνεται επίσης ένα σύστημα επενέργησης σε διάταξη τένοντα με νανοκαλώδια και γίνεται μελέτη της αντοχής του σε σχέση με τις μέγιστες δυνάμεις-ροπές που παρουσιάζονται κατά τον έλεγχο. Για την αναγνώριση των εξωτερικών δυνάμεων δοκιμάζονται διαφορετικά είδη εκτιμητών και εξετάζεται η απόδοσή τους στο συνολικό σύστημα. / The present PhD thesis has a key object the modeling and control of a micro robotic manipulator, represented by spherical particles in an environment with external Van Der Waals forces and taking friction into account. The main points are a) the insertion of the external forces in the micro robot model, b) the adaptive control used in order to follow a desired trajectory, with identification and cancellation of the external forces, the position identification and avoidance of obstacles in an unstructured micrometer scale environment and the trajectory planning towards a target point in the task space of the microrobot. Also a tendon like actuation system is proposed, using nanowires and its mechanical properties are studied in order to determine the viability of its use in relation to the required torques during the control process. For the external force identification scheme, various types of estimators are proposed and their efficiency in the system is studied.

This thesis is focused on developing vibration based terrain classification for an autonomous mining truck. The goal is to classify between good and bad gravel roads as well as good and bad asphalt roads. Current literature within vibration based terrain classification has been focused to a great extent on smaller research vehicles. On smaller research vehicles have roll-rate, pitch-rate and vertical acceleration been reported to yield the highest average classification rates. Common approaches for pre-processing the data consists of segmenting the data, apply filtering techniques, computing the Power Spectra Density (PSD), performing Principal Component Analysis (PCA) and compute the logarithms. How to do this specifically for an Autonomous Truck (AT) is not trivial. What signals from the trucks Internal Measurement Unit (IMU)s yields the highest average classification rates? How does one process the raw data in the best way, and what classification method performs the best for this for an AT? The AT studied here have five different IMUs that all measure ẍ, ÿ, z̈ acceleration, and ωroll, ωpitch, ωyaw rotational speed. One is located in the cab, and the other four are located in each of the four corners of the chassis. With these sensors empirical vibration data from different surfaces, speeds and loads was gathered with multiple identically equipped autonomous mining trucks. With this data were experiments conducted in order to find a high performing classifier that also was possible to implement in the ATs software in C++. The different signals were ranked according to the highest classification score, and different pre-processing parameters combined with different classification methods likewise were. ωyaw and ωpitch from the cab IMU, and z̈ from the rear right IMU were the ones that yielded the highest average classification rates. The pre-processing consists of segmenting the data, multiplying the segment with a window function, compute the one-sided PSD, logarithmize the PSD values and lastly normalize the data. A bagged classifier based on Support Vector Machine (SVM) with a Radial Basis Function (RBF) kernel showed the highest classification performance. The final multiclass classifier was a combination of three of these bagged classifiers in a tree structure. The F-measure rates for the four classes were {0.946, 0.98, 0.714, 0.879}. / Denna uppsats är fokuserad på att utveckla en vibrationsbaserad terrängigenkänningsalgoritm för en automatiserad tung lastbil som kommer att framföras i ojämn terräng, som ska klara av att känna igen bra och dåliga grusvägar, samt bra och dåliga asfaltsvägar. Befintlig litteratur inom området vibrationsbaserad terrängigenkänning har varit fokuserad i stor utsträckning på mindre forskningsfordon. På dessa är {ωrull, ωstigning, z̈} de signaler som resulterar i de högsta genomsnittliga korrekta terrängklassifikationerna. Befintliga förbearbetningmetoder för datan består i majoriteten av fallen av att segmentera och filtrera datan, beräkna spektrala effekttätheten (PSD) och logaritmera. Hur man gör detta är inte trivialt. Vilka signaler från lastbilens fem IMUer resulterar i det högsta prestandan för terrängigenkänning? Hur förarbetar man datan? Lastbilen studerad här har fem IMUer som har sex kanaler vardera, de mäter ẍ, ÿ, z̈ acceleration, och ωrull, ωstigning, ωgir rotationshastighet. En är placerad i lastbilens hytt och de andra fyra är placerade i varje hörn på chassit. Med dessa sensorer samlades vibrationsdata in på de fyra underlagen, med olika lastbilar, med olika last på flaket och med olika autonoma lastbilar, men som var konfigurerade på samma sätt. Experiment utfördes för att bestämma vilka signaler-, vilken förbearbetningsmetod på datan- samt vilken klassifieringsmetod som presterar bäst för den automatiserade lastbilen. Algoritmen var också anpassad för att vara möjlig att implementera i lastbilens mjukvara utan externa maskininlärnings bibliotek. De högst presterande signalerna var ωgir och ωstigning från IMUn i hytten, samt z̈ från IMUn monterad i chassits bakre högra hörn. Förbearbetningen bestod av att segmentera datasignalen, multiplicera den med en fönsterfunktion för att sedan beräkna den ensidiga spektrala effekttätheten (PSD), logaritmera alla värden och till slut normalisera datan. En stödvektormaskin (SVM) med en RBF kärna påvisade högst genomsnittliga klassifikationsresultat. Den slutgiltiga binära klassifieraren applicerade bagging för att förbättra prestandan genom att kombinera data från alla de tre högst presterande signalerna. Den slutgiltiga klassifieraren tränades på att skilja mellan de olika underlagen och var en kombination av tre bagged klassifierare i en trädstruktur. Prestandan med avseende på F-Measure för de fyra klasserna var {0.946, 0.98, 0.714, 0.879}.