Doctoral School

Reading date: 30/09/2025

• ESCUDERO RODRIGO, DIEGO: Dealing with the Anchoring Problem in Robotic Kitting using Behavior Trees
  
  Author: ESCUDERO RODRIGO, DIEGO
  Thesis file: (contact the Doctoral School to confirm you have a valid doctoral degree and to get the link to the thesis)
  Programme: DOCTORAL DEGREE IN AUTOMATIC CONTROL, ROBOTICS AND VISION
  Department: Department of Automatic Control (ESAII)
  Mode: Normal
  Deposit date: 01/09/2025
  Reading date: pending
  Reading time: pending
  Reading place: pending
  Thesis director: ARANDA LÓPEZ, JUAN | ALQUEZAR MANCHO, RENATO
  Thesis abstract: Industrial flexible manufacturing is a production method designed to adapt quickly to changes in product variety and volume. This is enabled by robots and perception systems capable of managing high-mix and high-volume production, ensuring quality and being easy to program. Flexible machines receive different types of parts using in-feed systems, and which types are used depends on the product that should be built. An important step is to provide to these machines all the parts required to be used during the production; this step is named as robotic kitting. Robotic kitting means the creation of parts assortment to be used later and these parts are selected from one or more containers in which there are different types of them randomly distributed. The procedure involves gathering the different parts to be moved and placed in a preparation area. In this area, the different parts are reassembled into a kit.The aim of this research was to develop an anchoring framework for Robotic `Intelligent' Kitting, a generalization of robotic kitting that incorporates symbols and their manipulation. Since symbols are required to solve this problem, the anchoring problem must be considered. This anchoring framework should enable generalist robots to anchor objects and actions required for a new task, through human-robot interaction and learning mechanisms. Our framework is based on automatically generated code and a human supervised approach, in which robot learning and human-robot interaction are used to anchor percepts and instructions to symbols during commissioning or reconfiguration phase. Genetic programming is a known technique for code generation that allows learning programs from scratch. So, it was combined with behavior trees for anchoring symbolic actions (e.g., find-part) to robot instructions at sensori-motor level (e.g., move-arm-to).An algorithm that combines genetic programming with conditional behavior trees (GP-CBT) is proposed. The core of the algorithm is composed by specific genetic operators, an evaluation criterion and the fallback swapper. As making easy the interaction with the operator is essential for our framework, this algorithm has been extended in order to generate action nodes automatically, allow the user to add task requirements, and update prior knowledge from previously learned tasks.
  

• PEDRAGOSA BATLLORI, GEMMA: Santa Coloma d'Andorra: el projecte d'una església a l'Andorra d'abans del S.XI.
  
  Author: PEDRAGOSA BATLLORI, GEMMA
  Thesis file: (contact the Doctoral School to confirm you have a valid doctoral degree and to get the link to the thesis)
  Programme: DOCTORAL DEGREE IN THEORY AND HISTORY OF ARCHITECTURE
  Department: Department of History and Theory of Architecture and Communication Techniques (THATC)
  Mode: Normal
  Deposit date: 01/09/2025
  Reading date: pending
  Reading time: pending
  Reading place: pending
  Thesis director: GRANELL TRIAS, ENRIQUE | GINER OLCINA, JOSEP
  Thesis abstract: The church of Santa Coloma d’Andorra belongs to one of the simplest and oldest architectural types of religious architecture: that of a single rectangular nave with a square apse. However, the simplicity of this type should not necessarily be associated with a straightforward or immediate construction or design.The aim of this study is to determine the extent to which the architecture of Santa Coloma follows a complex metrical design, which could only be achieved within a cultural context that, in Santa Coloma — located near two major cultural centres of the time, the Cathedral of La Seu d’Urgell and the Monastery of Sant Serni de Tavèrnoles — is highly plausible.In this work, architecture is used as archaeological material to analyse the key elements of the building’s architectural composition. Historiography has been reviewed, plans have been drawn up, the unit of measurement has been identified, and its dimensions studied in relation to the knowledge of proportion of the period and descriptions of biblical buildings. And it turns out that in order to conceive, design and build an apparently simple church like this, it was necessary to be familiar with the architecture represented in the Bible and with the arithmetical elaborations compiled by Boethius and Cassiodorus in the 6th century.We will therefore see a building which, although rural and seemingly modest, is the result of a layout and proportions based on a specific symbolic language, reflecting and documenting a body of knowledge and a way of applying it.
  

Reading date: 01/10/2025

• CHAVEZ AGUIRRE, JEAN PIERS NICOLAS: Post-Earthquake Functional Recovery and Resilience of Seismically Isolated Hospitals and Large-Scale Building Portfolios
  
  Author: CHAVEZ AGUIRRE, JEAN PIERS NICOLAS
  Thesis file: (contact the Doctoral School to confirm you have a valid doctoral degree and to get the link to the thesis)
  Programme: DOCTORAL DEGREE IN EARTHQUAKE ENGINEERING AND STRUCTURAL DYNAMICS
  Department: Department of Civil and Environmental Engineering (DECA)
  Mode: Normal
  Deposit date: 25/07/2025
  Reading date: 01/10/2025
  Reading time: 12:00
  Reading place: ETSECCPB. UPC, Campus Nord Sala Actes del Vèrtex Plaça d'Eusebi Güell, 608034 Barcelona
  Thesis director: LOPEZ ALMANSA, FRANCISCO | MURCIA DELSO, JUAN
  Thesis abstract: This dissertation presents a comprehensive contribution to assessing and optimizing the functional resilience of both individual hospital buildings and urban-scale building portfolios. The research addresses critical challenges in post-earthquake recovery by combining probabilistic methodologies, optimization algorithms, and machine learning techniques to evaluate and enhance the performance of hospital facilities, as well as assess large building portfolios with unprecedented computational efficiency. The first core contribution of this work lies in the development of a probabilistic framework for assessing the post-earthquake functionality of seismically isolated hospital buildings. Recognizing that traditional assessment methods and resilience metrics overlook complex interdependencies among non-structural components, medical equipment, and utilities, this study employs Bayesian Networks (BNs) to model such dependencies explicitly. The proposed framework quantifies the probabilistic damage of over 60 equipment and component types based on ground motion parameters, and propagates their impact through critical hospital departments. The framework allows for the direct computation of functionality indices using damage outputs.Building upon the proposed functionality loss model, the research also introduces a multiobjective optimization framework for hospital recovery planning using the NSGA-II genetic algorithm. The framework addresses the challenge of optimal labor allocation by minimizing both repair time and cost. It enables decision-makers to define the number of workers per repair group based on the importance of each component, which is classified according to its contribution to total functionality loss. The solution space is explored through a high-dimensional search of work effort allocations, and Pareto-optimal repair plans are obtained. The results indicate that excessive labor deployment yields diminishing returns in repair time while significantly increasing cost, with differences in repair time under design basis earthquake and maximum considered earthquake remaining below 13%.To address the scalability of resilience assessments at the urban scale, the dissertation proposes the Urban Cluster Earthquake Resilience (UCER) framework, which leverages unsupervised machine learning algorithms for regional seismic risk and resilience evaluation. Applied to a dataset of 23420 reinforced concrete and masonry buildings in a synthetic urban environment in Italy, the framework employs t-distributed stochastic neighbor embedding (t-SNE) for dimensionality reduction, HDBSCAN for density-based clustering, and K-Medoids for robust cluster representation. This process reduces the analysis complexity from thousands of buildings to 398 representative clusters. The results and findings of this dissertation highlight the benefits of integrated, data-driven approaches for assessing seismic resilience. By combining probabilistic interdependence modeling, artificial intelligence, and urban clustering, this research bridges the gap between detailed building-level assessments and large-scale regional resilience planning. Ultimately, the findings support the development of more resilient healthcare infrastructures and cities by enabling decision-makers to rely on scientifically grounded tools to prepare, respond, and recover effectively from seismic disasters.
  

• COLLAO LAZO, JORGE ALEJANDRO: Application of BIM visual programming algorithms for infrastructure projects
  
  Author: COLLAO LAZO, JORGE ALEJANDRO
  Thesis file: (contact the Doctoral School to confirm you have a valid doctoral degree and to get the link to the thesis)
  Programme: DOCTORAL DEGREE IN CONSTRUCTION ENGINEERING
  Department: Department of Civil and Environmental Engineering (DECA)
  Mode: Normal
  Deposit date: 29/07/2025
  Reading date: 01/10/2025
  Reading time: 16:00
  Reading place: Street Jordi Girona 1-3. C1 building, Room C1-007
  Thesis director: LOZANO GALANT, JOSE ANTONIO | TURMO CODERQUE, JOSE
  Thesis abstract: The BIM digitization has generated a growing automation of traditional AECO project development processes. However, this automation has mainly benefited building projects, which has generated a critical gap in infrastructure projects, such as (1) bridges, (2) roads, and (3) tunnels. This gap is explained by infrastructure projects' lack of standardization compared with building projects. Recently, the BIM industry has incorporated conventional computer programming as a tool capable of partially reducing this lack of automation, developing code-line algorithms. However, both the work interface and the high technical skills required to produce these scripts have been unfriendly to AECO industry professionals. To resolve this gap, BIM software development companies generated an alternative algorithm creation technique called Visual Programming (VP). This technique creates their scripts through visual expressions, represented as process charts instead of the code lines of conventional programming, optimizing the working interface of different human resources from AECO projects. However, the development of VP algorithms for infrastructure projects is still limited.To fill these gaps, the objective of this thesis will be to study the potential automation of infrastructure project processes, developing an integrated platform between VP algorithms and BIM models.At a general level, the platform proposed by this thesis has the following 2 elements: (1) VP algorithms; These custom scripts have the specific function of processing information from external databases and deliver specific data as outputs, and (2) BIM models; These information models will digitally collect the outputs of the VP algorithms and associate them with specific BIM families of the infrastructure projects under study. This thesis focuses on studying transversal problems of Civil Engineering using VP-BIM tools through the development of the following applications:Firstly, this thesis develops a calculation module to estimate (1) greenhouse (GHG) emissions; these pollutants contribute to global warming and climate change by trapping heat in the Earth's atmosphere [1–3] and (2) carbon footprint for vehicle traffic on specific roads in any country with the appropriate traffic and vehicle fleet data. The main novelty of this module is the automated integration of the GHG emission factors recommended by the European Environmental Agency (EEA) TIER 1 level with specific fleet data through a customized VP script. This proposed module was applied on a road in Barcelona (Spain), and the results were compared with a similar study made by the Barcelona transport agency. In addition, specific calculation modules were developed to measure the impact of emission reduction strategies.The second application developed by this thesis aims to use visual programming as an educational tool. A parametric programming workflow is employed to replicate the structural behavior of a beam within a BIM Model. This approach enables students to visualize and analyze structural performance, fostering a deeper connection between classroom theory and hands-on experimentation. This system aims to improve the visualization and understanding of structural data for AEC industry professionals by generating dynamic 3D and 2D models. The educational impact of this tool was assessed through a survey conducted in a structural analysis course at the University of La Serena (Chile), demonstrating its effectiveness in enhancing student comprehension and engagement with modern engineering practices.
  

• DELGADO GUERRERO, JUAN ANTONIO: Learning latent structures for robotic assistance in daily manipulation tasks
  
  Author: DELGADO GUERRERO, JUAN ANTONIO
  Thesis file: (contact the Doctoral School to confirm you have a valid doctoral degree and to get the link to the thesis)
  Programme: DOCTORAL DEGREE IN AUTOMATIC CONTROL, ROBOTICS AND VISION
  Department: Institute of Robotics and Industrial Informatics (IRI)
  Mode: Normal
  Deposit date: 02/09/2025
  Reading date: pending
  Reading time: pending
  Reading place: pending
  Thesis director: TORRAS GENIS, CARMEN | COLOMÉ FIGUERAS, ADRIÀ
  Thesis abstract: Robotic domestic assistance presents significant challenges due to the complexity of modeling everyday manipulation tasks, especially those involving deformable objects like cloth. Traditional approaches often struggle with high-dimensional state representations, dynamic uncertainties, and the need for safe human-robot interaction. This thesis addresses these challenges by developing novel machine learning methods based on latent variable models to enable efficient, adaptive, and safe robotic manipulation.First, we propose a Gaussian Process Latent Variable Model (GPLVM) framework combined with Bayesian Optimization (BO) to learn high-dimensional robot motion policies with minimal data. This approach reduces the parameter space dimensionality while preserving task-relevant features, achieving faster convergence than other existing model-free alternatives.Next, we extend this framework to contextual learning using Covariate GPLVM (c-GPLVM), allowing robots to adapt to environmental changes (e.g., user preferences, object positions) without retraining. Experiments in feeding and shoe-fitting tasks demonstrate improved generalization with fewer samples compared to state-of-the-art contextual policy search methods.For dynamic cloth manipulation, we introduce the Controlled Gaussian Process Dynamical Model (CGPDM), which embeds control actions into a low-dimensional latent space to predict cloth motion under robot manipulation. Evaluations in simulated and real-world bimanual cloth handling show that CGPDM accurately generalizes to unseen actions, even with limited training data.Finally, we address safety in human-robot interaction by proposing Cartesian control enhancements for redundant manipulators, including error saturation, singularity avoidance, and impedance tuning. These measures mitigate risks during physical interaction, ensuring stable and compliant robot behavior.Together, these contributions advance robotic cloth manipulation by combining data-efficient learning, context-aware adaptation, and safe control, paving the way for practical deployment in assistive and household robotics.