Structure learning of graphical models for task-oriented robot grasping

In the collective imaginaries a robot is a human like machine as any androids in science fiction. However the type of robots that you will encounter most frequently are machinery that do work that is too dangerous, boring or onerous. Most of the robots in the world are of this type. They can be found in auto, medical, manufacturing and space industries. Therefore a robot is a system that contains sensors, control systems, manipulators, power supplies and software all working together to perform a task. The development and use of such a system is an active area of research and one of the main problems is the development of interaction skills with the surrounding environment, which include the ability to grasp objects. To perform this task the robot needs to sense the environment and acquire the object informations, physical attributes that may influence a grasp. Humans can solve this grasping problem easily due to their past experiences, that is why many researchers are approaching it from a machine learning perspective finding grasp of an object using information of already known objects. But humans can select the best grasp amongst a vast repertoire not only considering the physical attributes of the object to grasp but even to obtain a certain effect. This is why in our case the study in the area of robot manipulation is focused on grasping and integrating symbolic tasks with data gained through sensors. The learning model is based on Bayesian Network to encode the statistical dependencies between the data collected by the sensors and the symbolic task. This data representation has several advantages. It allows to take into account the uncertainty of the real world, allowing to deal with sensor noise, encodes notion of causality and provides an unified network for learning. Since the network is actually implemented and based on the human expert knowledge, it is very interesting to implement an automated method to learn the structure as in the future more tasks and object features can be introduced and a complex network design based only on human expert knowledge can become unreliable. Since structure learning algorithms presents some weaknesses, the goal of this thesis is to analyze real data used in the network modeled by the human expert, implement a feasible structure learning approach and compare the results with the network designed by the expert in order to possibly enhance it.

Potential effect of marine traffic on white-beaked dolphin (Lagenorhynchus albirostris) distribution and behaviour in Faxaflói Bay, southwest Iceland

Faxaflói bay is a short, wide and shallow bay situated in the southwest of Iceland. Although hosting a rather high level of marine traffic, this area is inhabited by many different species of cetaceans, among which the white-beaked dolphin (Lagenorhynchus albirostris), found here all year-round. This study aimed to evaluate the potential effect of increasing marine traffic on white-beaked dolphins distribution and behaviour, and to determine whether or not a variation in sighting frequencies have occurred throughout years (2008 – 2014). Data on sightings and on behaviour, as well as photographic one, has been collected daily taking advantage of the whale-watching company “Elding” operating in the bay. Results have confirmed the importance of this area for white-beaked dolphins, which have shown a certain level of site fidelity. Despite the high level of marine traffic, this dolphin appears to tolerate the presence of boats: no differences in encounter durations and locations over the study years have occurred, even though with increasing number of vessels, an increase in avoidance strategies has been displayed. Furthermore, seasonal differences in probabilities of sightings, with respect to the time of the day, have been found, leading to suggest the existence of a daily cycle of their movements and activities within the bay. This study has also described a major decline in sighting rates throughout years raising concern about white-beaked dolphin conservation status in Icelandic waters. It is therefore highly recommended a new dedicated survey to be conducted in order to document the current population estimate, to better investigate on the energetic costs that chronic exposure to disturbances may cause, and to plan a more suitable conservation strategy for white-beaked dolphin around Iceland.

Mission synthesis of sine-on-random excitations for accelerated vibration qualification testing

In most real-life environments, mechanical or electronic components are subjected to vibrations. Some of these components may have to pass qualification tests to verify that they can withstand the fatigue damage they will encounter during their operational life. In order to conduct a reliable test, the environmental excitations can be taken as a reference to synthesize the test profile: this procedure is referred to as “test tailoring”. Due to cost and feasibility reasons, accelerated qualification tests are usually performed. In this case, the duration of the original excitation which acts on the component for its entire life-cycle, typically hundreds or thousands of hours, is reduced. In particular, the “Mission Synthesis” procedure lets to quantify the induced damage of the environmental vibration through two functions: the Fatigue Damage Spectrum (FDS) quantifies the fatigue damage, while the Maximum Response Spectrum (MRS) quantifies the maximum stress. Then, a new random Power Spectral Density (PSD) can be synthesized, with same amount of induced damage, but a specified duration in order to conduct accelerated tests. In this work, the Mission Synthesis procedure is applied in the case of so-called Sine-on-Random vibrations, i.e. excitations composed of random vibrations superimposed on deterministic contributions, in the form of sine tones typically due to some rotating parts of the system (e.g. helicopters, engine-mounted components, …). In fact, a proper test tailoring should not only preserve the accumulated fatigue damage, but also the “nature” of the excitation (in this case the sinusoidal components superimposed on the random process) in order to obtain reliable results. The classic time-domain approach is taken as a reference for the comparison of different methods for the FDS calculation in presence of Sine-on-Random vibrations. Then, a methodology to compute a Sine-on-Random specification based on a mission FDS is presented.