Search

This collaborative study with Wind Power Lab and IRI aimed to enhance image segmentation algorithms, specifically region growing, through innovative image pre-processing techniques. The core objective was to modify the color space of images using statistical methods and optimization tools to improve algorithm performance. The project involved the development and analysis of the region growing algorithm. We explored various colour scales and assessed their impact on segmentation outcomes, leading to a deeper understanding of customised image colour scale modifications. The pivotal phase of this study involved formulating an optimization problem that targeted a linear constant transformation of image colours. The aim was to improve windmill identification against varied backgrounds. This was achieved by minimizing the distances between pixels associated with windmills while maximizing background pixel distances. The approach used was statistically analogous to binary classification problems. Analytical and experimental methods, including gradient descent, were used to define a linear transformer vector for modifying colour channels. Despite challenges in transforming colour space into three dimensions and limited testing data, the algorithm demonstrated improved segmentation in three dimensions. This suggests that it is effective in preprocessing before segmentation. The study concluded that while colour space transformation can enhance image segmentation, the seeded region growing algorithm showed superior results.

In this work, we introduce a collision model specifically tailored for the simulation of inextensible textiles. The model considers friction, contacts, and inextensibility constraints all at the same time without any decoupling. Self-collisions are modeled in a natural way that allows considering the thickness of cloth without introducing unwanted oscillations. The discretization of the equations of motion leads naturally to a sequence of quadratic problems with inequality and equality constraints. In order to solve these problems efficiently, we develop a novel active-set algorithm that takes into account past active constraints to accelerate the resolution of unresolved contacts. We put to a test the developed collision procedure with diverse scenarios involving static and dynamic friction, sharp objects, and complex-topology folding sequences. Finally, we perform an experimental validation of the collision model by comparing simulations with recordings of real textiles as given by a Motion Capture System. The results are very accurate, having errors around 1 cm for DIN A2 textiles (42 x 59.4 cm) even in difficult scenarios involving fast and strong hits with a rigid object., Peer Reviewed, Postprint (published version)

In this paper we introduce SMPLicit, a novel generative model to jointly represent body pose, shape and clothing geometry; and LayerNet, a deep network that given a single image of a person simultaneously performs detailed 3D reconstruction of body and clothes. In contrast to existing learning-based approaches that require training specific models for each type of garment, SMPLicit can represent in a unified manner different garment topologies (eg from sleeveless tops to hoodies and open jackets), while controlling other properties like garment size or tightness/looseness. LayerNet follows a coarse-to-fine multi-stage strategy by first predicting smooth cloth geometries from SMPLicit, which are then refined by an image-guided displacement network that gracefully fits the body recovering high-frequency details and wrinkles. LayerNet achieves competitive accuracy in the task of 3D reconstruction against current `garment-agnostic' state of the art for images of people in up-right positions and controlled environments, and consistently surpasses these methods on challenging body poses and uncontrolled settings. Furthermore, the semantically rich outcome of our approach is suitable for performing Virtual Try-on tasks directly on 3D, a task which, so far, has only been addressed in the 2D domain., This work is supported in part by the Spanish government with the project MoHuCo PID2020-120049RBI00, the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) - 409792180 (Emmy Noether Programme, project: Real Virtual Humans) and German Federal Ministry of Education and Research (BMBF): Tubingen AI ¨Center, FKZ: 01IS18039A. Gerard Pons-Moll is a member of the Machine Learning Cluster of Excellence, EXC number 2064/1 – Project number 390727645. The project was made possible by funding from the Carl Zeiss Foundation. We thank NVIDIA for donating GPUs, Peer Reviewed, Postprint (author's final draft)

Frailty is a crucial indicator in determining the well-being of older adults in terms of their health. With the growing number of elderly people, the demand for geriatricians is increasing, which means that they have less time to spend with each patient. The current methods for frailty assessment use simple tests that are time-consuming and do not require specific medical expertise. To address this issue, this paper proposes the use of social robots to assess frailty autonomously. It presents a practical proposal that defines the robot’s behavior and explains the design and implementation concepts. Finally, it discusses some of the challenges that may arise from introducing social robots as frailty evaluators., This work was supported by the project ROB-IN PLEC2021-007859 funded by MCIN/ AEI /10.13039/501100011033 and by the "European Union NextGenerationEU/PRTR"; project CHLOE-GRAPH PID2020-118649RB-I00 funded by MCIN/AEI /10.13039/501100011033; and the 23S06141-001 FRAILWATCH project funded by Barcelona Ciencia 2020-2023 Plan., Peer Reviewed, Postprint (published version)

Realistic physics engines play a crucial role for learning to manipulate deformable objects such as garments in simulation. By doing so, researchers can circumvent challenges such as sensing the deformation of the object in the real-world. In spite of the extensive use of simulations for this task, few works have evaluated the reality gap between deformable object simulators and real-world data. We present a benchmark dataset to evaluate the sim-to-real gap in cloth manipulation. The dataset is collected by performing a dynamic as well as a quasi-static cloth manipulation task involving contact with a rigid table. We use the dataset to evaluate the reality gap, computational time, and simulation stability of four popular deformable object simulators: MuJoCo, Bullet, Flex, and SOFA. Additionally, we discuss the benefits and drawbacks of each simulator. The benchmark dataset is open-source. Supplementary material, videos, and code, can be found at https://sites.google.com/view/cloth-sim2real-benchmark., This work was supported in part by Academy of Finland under Grant 317020, in part by Business Finland (decision 9249/31/2021), in part by EU’s H2020 Programme through the CLOTHILDE Project under Grant ERC-2016-ADG-741930, in part by EU’s Horizon Europe Programme through the SoftEnable Project under Grant Agreement 101070600, and in part by MCIN/AEI/10.13039/501100011033 through CHLOE-GRAPH Project under Grant PID2020-118649RB-I00., Peer Reviewed, Postprint (published version)

This paper addresses the post-fault selection of an actuators configuration for flow-based networks with back-up components. The proposed reconfiguration methodology consists of an offline and an online phase. On the one hand, an offline analysis looks for the minimal configurations for which the economic cost of the (best) steady-state trajectory that can be achieved using a robust model predictive control (MPC) policy is admissible. On the other hand, at fault detection time, an online search for the best actuators configuration to cope with the transient induced by the fault is conducted in the superset of each minimal configuration calculated offline. With this strategy, the final new configuration is computed by sequentially solving a set of mixed-integer programs whose constraints are derived from single-layer robust MPC schemes coupled with local controllers designed for the a priori minimal configurations identified offline. A portion of a water transport network is used to show the performance the proposed solution., Peer Reviewed, Postprint (author's final draft)

This article presents a novel response methodology for the operational recovery of a drinking water network after an incident causes an interruption of service. The proposed optimization-based methodology allows computing the optimal set of interventions to be performed in order to mitigate, or even prevent, the impact of the incident on the network operation. Besides, a proof-of-concept scheme has been designed for the automatic generation of failure scenarios and the systematic implementation and validation of the proposed response methodology. Several results are presented to demonstrate the capability of the methodology to mitigate harmful incidents, as well as the performance improvements derived from the application of the obtained interventions., Peer Reviewed, Postprint (author's final draft)

© 2023 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting /republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works, In the future, robots are expected to autonomously interact and/or collaborate with humans, who will increase the uncertainty during the execution of tasks, provoking online adaptations of robots' plans. Hence, trustworthy robots must be able to store, retrieve and narrate important knowledge about their collaborations and adaptations. In this article, it is proposed a sound methodology that integrates three main elements. First, an ontology for collaborative robotics and adaptation to model the domain knowledge. Second, an episodic memory for time-indexed knowledge storage and retrieval. Third, a novel algorithm to extract the relevant knowledge and generate textual explanatory narratives. The algorithm produces three different types of outputs, varying the specificity, for diverse uses and preferences. A pilot study was conducted to evaluate the usefulness of the narratives, obtaining promising results. Finally, we discuss how the methodology can be generalized to other ontologies and experiences. This work boosts robot explainability, especially in cases where robots need to narrate the details of their short and long-term past experiences., Peer Reviewed, Postprint (author's final draft)

This work studies the impact of morphological symmetries in learning applications in robotics. Morphological symmetries are a predominant feature in both biological and robotic systems, arising from the presence of planes/axis of symmetry in the system's morphology. This results in harmonious duplication and distribution of body parts (e.g., humans' sagittal/left-right symmetry). Morphological symmetries become a significant learning prior as they extend to symmetries in the system's dynamics, optimal control policies, and in all proprioceptive and exteroceptive measurements, related to the system's dynamics evolution \cite{ordonez2023discrete}. Exploiting these symmetries in learning applications offers several advantageous outcomes, such as the use of data augmentation to mitigate the cost and challenges of data collection, or the use of equivariant/invariant function approximation models (e.g., neural networks) to improve sample efficiency and generalization, while reducing the number of trainable parameters. Lastly, we provide a video presentation and an open access repository reproducing our experiments and allowing for rapid prototyping in robot learning applications exploiting morphological symmetries., This work is supported by the Spanish government with the project MoHuCo PID2020-120049RB-I00 and the ERA-Net Chistera project IPALM PCI2019-103386., Peer Reviewed, Postprint (published version)

© 2023 by the authors, Several clinical trials are working on drug development for Duchenne and Becker muscular dystrophy (DMD and BMD) treatment, and, since the expected increase in dystrophin is relatively subtle, high-sensitivity quantification methods are necessary. There is also a need to quantify dystrophin to reach a definitive diagnosis in individuals with mild BMD, and in female carriers. We developed a method for the quantification of dystrophin in DMD and BMD patients using spectral confocal microscopy. It offers the possibility to capture the whole emission spectrum for any antibody, ensuring the selection of the emission peak and allowing the detection of fluorescent emissions of very low intensities. Fluorescence was evaluated first on manually selected regions of interest (ROIs), proving the usefulness of the methodology. Later, ROI selection was automated to make it operator-independent. The proposed methodology correctly classified patients according to their diagnosis, detected even minimal traces of dystrophin, and the results obtained automatically were statistically comparable to the manual ones. Thus, spectral imaging could be implemented to measure dystrophin expression and it could pave the way for detailed analysis of how its expression relates to the clinical course. Studies could be further expanded to better understand the expression of dystrophin-associated protein complexes (DAPCs)., This research was partially founded by “Somriures Valents” (private grant)., Peer Reviewed, Postprint (published version)

© 2023 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works., In this paper, we propose a novel fully unsupervised framework that learns action representations suitable for the action segmentation task from the single input video itself, without requiring any training data. Our method is a deep metric learning approach rooted in a shallow network with a triplet loss operating on similarity distributions and a novel triplet selection strategy that effectively models temporal and semantic priors to discover actions in the new representational space. Under these circumstances, we successfully recover temporal boundaries in the learned action representations with higher quality compared with existing unsupervised approaches. The proposed method is evaluated on two widely used benchmark datasets for the action segmentation task and it achieves competitive performance by applying a generic clustering algorithm on the learned representations., This work was supported by the project PID2019-110977GA-I00 funded by MCIN/ AEI/ 10.13039/501100011033 and by ”ESF Investing in your future”, Peer Reviewed, Postprint (author's final draft)

In this chapter we look at science fiction from a perspective that goes beyond pure entertainment. Such literary gender can play an important role in bringing science closer to society by helping to popularize scientific knowledge and discoveries while engaging the public in debates which, in turn, can help direct scientific development towards building a better future for all. Written based on a tutorial given by the first author at ACAI 2021, this chapter addresses, in its first part, how science and science fiction can inspire each other and, in its second part, how science fiction can be used as an educational tool in teaching ethics of AI and robotics. Each of the two parts is supplemented with sections containing the questions asked by the audience during the tutorial as well as the provided answers., Supported by the European Research Council (ERC) within the European Union Horizon 2020 Programme under grant agreement ERC–2016–ADG–741930 (CLOTHILDE: CLOTH manIpulation Learning from DEmonstrations), Peer Reviewed, Postprint (author's final draft)

We present a comprehensive study on discrete morphological symmetries of dynamical systems, which are commonly observed in biological and artificial locomoting systems, such as legged, swimming, and flying animals/robots/virtual characters. These symmetries arise from the presence of one or more planes/axis of symmetry in the system’s morphology, resulting in harmonious duplication and distribution of body parts. Significantly, we characterize how morphological symmetries extend to symmetries in the system’s dynamics, optimal control policies, and in all proprioceptive and exteroceptive measurements related to the system’s dynamics evolution. In the context of data-driven methods, symmetry represents an inductive bias that justifies the use of data augmentation or symmetric function approximators. To tackle this, we present a theoretical and practical framework for identifying the system’s morphological symmetry group G and characterizing the symmetries in proprioceptive and exteroceptive data measurements. We then exploit these symmetries using data augmentation and G-equivariant neural networks. Our experiments on both synthetic and real-world applications provide empirical evidence of the advantageous outcomes resulting from the exploitation of these symmetries, including improved sample efficiency, enhanced generalization, and reduction of trainable parameters., This work is supported by the Spanish government with the project MoHuCo PID2020-120049RB-I00 and the ERA-Net Chistera project IPALM PCI2019-103386., Peer Reviewed, Postprint (author's final draft)

L'augment del comerç electrònic en les grans ciutats durant els darrers anys ha presentat desafiaments importants, entre ells un increment del trànsit i de la contaminació. L’objectiu d’aquest treball és plantejar les bases pel disseny d’un sistema d’orientació d’un robot repartidor de paquets a través de comandes de veu, que ajudaria a reduir el nombre de vehicles als carrers i carreteres. El prototip del sistema de localització del so consisteix de dos micròfons electret amb amplificació connectats a un Arduino Uno. Les dades provinents d’una font de so puntual s’han mostrejat i treballat amb la combinació dels programes Arduino IDE i Matlab. S’ha treballat amb la hipòtesis que el micròfon més proper a la font ha de detectar un volum de so superior respecte al segon. D’aquesta manera, mitjançant la presa de mesures a diferents angles, s’aconseguirà un model matemàtic que ens permetrà determinar la posició de la font acústica. Un cop dissenyada i realitzada la part experimental del projecte, hem determinat que els resultats obtinguts no han sigut els esperats, ja que s’havia subestimat l’efecte de reverberació del so, és a dir, l’entorn físic tenia un impacte inevitable a l’hora d’enregistrar i comparar el volum sonor. De forma alternativa, s’ha mantingut el prototip i el mètode d’enregistrament de les dades, però s’ha canviat l’enfocament de l’estudi d’aquestes. En comptes de contrastar el volum enregistrat pels dos micròfons, s’ha decidit comparar la diferència de temps d’arribada del so als dos destins, El aumento del comercio electrónico en las grandes ciudades durante los últimos años ha presentado desafíos importantes, entre ellos un incremento del tráfico y de la contaminación. El objetivo de este trabajo es sentar las bases para el diseño de un sistema de orientación de un robot repartidor de paquetes a través de comandos de voz, que ayudaría a reducir el número de vehículos en las calles y carreteras. El prototipo del sistema de localización del sonido consiste en dos micrófonos electret con amplificación conectados a un Arduino Uno. Los datos provenientes de una fuente de sonido puntual se han muestreado y procesado con la combinación de los programas Arduino IDE y Matlab. Se ha trabajado con la hipótesis de que el micrófono más cercano a la fuente debe detectar un volumen de sonido superior al segundo. De esta manera, mediante la toma de medidas a diferentes ángulos, se logrará un modelo matemático que nos permitirá determinar la posición de la fuente acústica. Una vez diseñada y realizada la parte experimental del proyecto, hemos determinado que los resultados obtenidos no han sido los esperados, ya que se había subestimado el efecto de la reverberación del sonido, es decir, el entorno físico tenía un impacto inevitable a la hora de registrar y comparar el volumen sonoro. De forma alternativa, se ha mantenido el prototipo y el método de registro de los datos, pero se ha cambiado el enfoque del estudio de estos. En lugar de contrastar el volumen registrado por los dos micrófonos, se ha decidido comparar la diferencia de tiempo de llegada del sonido a los dos destinos, The proliferation of e-commerce in major cities over the course of recent years has presented significant challenges, including increases in traffic and pollution. The objective of this work is to lay the design foundations for a voice-commanded orientation system intended to be used by a package delivery robot, which would, thus, help in reducing the number of vehicles on the streets and roads. The prototype of the sound localization system consists of two electret microphones with amplification connected to an Arduino Uno. Data from a point source of sound was sampled and processed using a combination of Arduino IDE and Matlab programs. We have worked with the hypothesis that the microphone closest to the source should detect a higher sound volume than the second one. In this way, by taking measurements at different angles, we will achieve a mathematical model with which we can determine the position of the acoustic source. Once the experimental part of the project was designed and carried out, we determined that the results obtained were, unfortunately, not as expected. This was due to a sub-estimation of the effect produced by sound reverberation, meaning, therefore, that the physical environment had an inevitable impact on recording and comparing sound volume. Alternatively, the prototype and the data recording method have been retained, but the focus of the study has changed. Instead of contrasting the volume recorded by the two microphones, a decision was taken to compare the difference in sound arrival time at the two destinations

La robótica asistencial es un campo en rápido crecimiento destinado a ayudar a los cuidadores en hospitales, centros de rehabilitación y residencias de ancianos, así como a empoderar a las personas con movilidad reducida en el hogar para que puedan vivir de manera autónoma. Este es un ámbito altamente interdisciplinar, que requiere aunar no solo conocimientos de varias áreas tecno-científicas, sino también de ciencias sociales y humanidades. Es crucial que en la concepción, desarrollo y despliegue de los asistentes robóticos participen todos los actores intervinientes en el sector de la asistencia y los cuidados. El co-diseño de la interacción persona-robot es particularmente importante en este contexto, debido a la gran diversidad de condicionantes, tanto tecnológicos como éticos, a que está sujeta. Dicha interacción debe ser amigable, adaptativa y segura, tanto para el cuidador como para el paciente, al tiempo que debe preservar la dignidad y el control por parte de este último y no propiciar su aislamiento social, sino por el contrario estimular la comunicación y la realización de actividades con personas de su entorno próximo., El trabajo descrito se enmarca en el proyecto CLOTHILDE, financiado por el European Research Council (ERC) en el programa Horizonte 2020 de la Unión Europea (ERC Advanced Grant n.º 741930), así como en la Plataforma Temática del CSIC (PTI+ NeuroAging: Entender el envejecimiento desde la I+D+I), financiado por la Comisión Europea en el programa Next-GenerationEU, Postprint (author's final draft)

© The Author(s) 2023, Recent research in Human Robot Collaboration (HRC) has spread and specialised in many sub-fields. Many show considerable advances, but the human–robot collaborative navigation (HRCN) field seems to be stuck focusing on implicit collaboration settings, on hypothetical or simulated task allocation problems, on shared autonomy or on having the human as a manager. This work takes a step forward by presenting an end-to-end system capable of handling real-world human–robot collaborative navigation tasks. This system makes use of the Social Reward Sources model (SRS), a knowledge representation to simultaneously tackle task allocation and path planning, proposes a multi-agent Monte Carlo Tree Search (MCTS) planner for human–robot teams, presents the collaborative search as a testbed for HRCN and studies the usage of smartphones for communication in this setting. The detailed experiments prove the viability of the approach, explore collaboration roles adopted by the human–robot team and test the acceptability and utility of different communication interface designs., Open Access funding provided thanks to the CRUE-CSIC agreement with Springer Nature. This work was supported under the Spanish State Research Agency through the Maria de Maeztu Seal of Excellence to IRI (MDM-2016-0656) and ROCOTRANSP project (PID2019- 106702RB-C21 / AEI / 10.13039/501100011033), the European research grant TERRINet (H2020-INFRAIA-2017-1-730994) and by JST Moonshot R & D Grant Number JPMJMS2011-85., Peer Reviewed, Postprint (published version)

La població de més de 65 anys ha augmentat de manera progressiva en les darreres dècades i els estudis prospectius avisen que continuarà creixent de manera cada vegada més accelerada. La tecnologia pot contribuir tant a la prevenció com a la realització de tasques assistencials rutinàries. En particular, la robòtica assistencial destaca com un element decisiu per a construir un futur amb un envelliment saludable i sostenible, però cal que el seu desplegament es faci d’acord amb principis ètics, que sigui supervisat per òrgans competents i que la formació s’estengui a tots els agents implicats., El treball descrit s’emmarca en el projecte CLOTHILDE, finançat per l’European Research Council (ERC) dins del programa Horitzó 2020 de la Unió Europea (ERC Advanced Grant núm. 741930), així com en la Plataforma Temàtica del CSIC (PTI+ Neuro-Aging: Entendre l’envelliment des de la R+D+I), finançat per la Comissió Europea dins del programa NextGenerationEU., Postprint (published version)

© 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission., This brief contributes to the application of model predictive control (MPC) to address the combined sewer overflow (CSO) problem in urban drainage systems (UDSs) with uncertainty. In UDS, dealing with uncertainty in rain forecast and dynamic models is crucial due to the possible impact on the UDS control performance. Two different MPC approaches are considered: tube-based MPC (T-MPC) and chance-constrained MPC (CC-MPC), which represent uncertainty in deterministic and stochastic manners, respectively. This brief presents how to apply T-MPC to UDS, by establishing a mathematical relation with CC-MPC, and a rigorous mathematical comparison. Based on simulations using the Astlingen benchmark UDS, the strengths and weaknesses of the performance of T-MPC and CC-MPC in UDS were compared. Differences in the involved mathematical computations have also been analyzed. Moreover, the comparison in performance also indicates the applicability of each MPC approach in different uncertainty scenarios., This work was supported in part by the Supervision and fault-tolerant control of smart infrastructures based on advanced learning and optimization (L-BEST) Project of the Spanish Agency of Research under Grant PID2020-115905RB-C21 and in part by the Real-time pollution-based control of urban drainage and sanitation systems for protection of receiving waters (RUBIES) Project of the European Commission under Grant LIFE20 ENV/FR/000179, Peer Reviewed, Postprint (author's final draft)

We study the perception problem for garments (e.g. a pair of pants) using tools from computational topology: the identification of their geometry and position from point-cloud samples, as obtained e.g. with 3D scanners. We present a reconstruction algorithm based on Morse theory that proceeds directly from the point-cloud to obtain a cellular decomposition of the surface derived via a Morse function. No intermediate triangulation or local implicit equations are used, avoiding reconstruction-induced artifices. The results are a piecewise parametrization of the surface as a union of Morse cells, suitable for tasks such as noise-filtering or mesh-independent reparametrization, and a cell complex of small rank determining the surface topology. This algorithm can be applied to smooth surfaces with or without boundary, embedded in an ambient space of any dimension., Peer Reviewed, Postprint (published version)

© 2023 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting /republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works, Robotic cloth manipulation is an increasingly relevant area of research, challenging classic control algorithms due to the deformable nature of cloth. While it is possible to apply linear model predictive control to make the robot move the cloth according to a given reference, this approach suffers from a large dimensionality of the state-space representation of the cloth models. To address this issue, in this work we study the application of an input-output model predictive control strategy, based on quadratic dynamic matrix control, to robotic cloth manipulation. To account for uncertain disturbances on the cloth's motion, we further extend the algorithm with suitable chance constraints. In extensive simulated experiments, involving disturbances and obstacle avoidance, we show that quadratic dynamic matrix control can be successfully applied in different cloth manipulation scenarios, with significant gains in optimization speed compared to standard model predictive control strategies. The experiments further demonstrate that the closed-loop model used by quadratic dynamic matrix control can be beneficial to the tracking accuracy, leading to improvements over the standard predictive control strategy. Moreover, a preliminary experiment on a real robot shows that quadratic dynamic matrix control can indeed be employed in real settings., This work was supported by the project CLOTHILDE (“CLOTH manIpulation Learning from DEmonstrations”), funded by the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (Advanced Grant agreement No 741930)., Peer Reviewed, Postprint (author's final draft)

This paper presents an approach to ensure conditions on Variable Impedance Controllers through the off-line tuning of the parameters involved in its description. In particular, we prove its application to term modulations defined by a Learning from Demonstration technique. This is performed through the assessment of conditions regarding safety and performance, which encompass heuristics and constraints in the form of Linear Matrix Inequalities. Latter ones allow to define a convex optimisation problem to analyse their fulfilment, and require a polytopic description of the VIC, in this case, obtained from its formulation as a discrete-time Linear Parameter Varying system. With respect to the current state-of-art, this approach only limits the term definition obtained by the Learning from Demonstration technique to be continuous and function of exogenous signals, i.e. external variables to the robot. Therefore, using a solution-search method, the most suitable set of parameters according to assessment criteria can be obtained. Using a 7-DoF KinovaGen3 manipulator, validation and comparison against solutions with relaxed conditions are performed. The method is applied to generate Variable Impedance Controllers for a pulley belt looping task, inspired by the Assembly Challenge for Industrial Robotics in World Robot Summit 2018, to reduce the exerted force with respect to a standard (constant) Impedance Controller. These controllers fulfil a set of safety constraints, namely stability, bounds on task variables and maximum response overshooting; and their performance is determined by the User Preference heuristic, which allows to intuitively define the desired compliant behaviour along the task. In the context of the task, this is used to generate new controllers for one-off modifications of the nominal belt looping task setup without new demonstrations., This work is partially supported by MCIN/ AEI /10.13039/501100011033 and by the ”European Union NextGenerationEU/PRTR” under the project ROB-IN (PLEC2021-007859); and by MCIN/ AEI /10.13039/ 501100011033, Spain, under the project CHLOE-GRAPH (PID2020- 119244GB-I00). Authors also want to thank Adriá Colomé and Edoardo Caldarelli for their comments and help throughout this work., Peer Reviewed, Postprint (published version)

© 2023 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting /republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works, With the relentless growth of the wind industry, there is an imperious need to design automatic data-driven solutions for wind turbine maintenance. As structural health monitoring mainly relies on visual inspections, the first stage in any automatic solution is to identify the blade region on the image. Thus, we propose a novel segmentation algorithm that strengthens the U-Net results by a tailored loss, which pools the focal loss with a contiguity regularization term. To attain top performing results, a set of additional steps are proposed to ensure a reliable, generic, robust and efficient algorithm. First, we leverage our prior knowledge on the images by filling the holes enclosed by temporarily-classified blade pixels and by the image boundaries. Subsequently, the mislead classified pixels are successfully amended by training an on-the-fly random forest. Our algorithm demonstrates its effectiveness reaching a non-trivial 97.39% of accuracy., This work has been supported by the Innovation Fund Denmark under 2021 ID1044-0044A and by the project MoHuCo PID2020-120049RB-I00 funded by MCIN/AEI/10.13039/501100011033., Peer Reviewed, Postprint (author's final draft)

© 2023 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works., The introduction of robotic technology in an existing scenario must be analyzed from the point of view of all the human roles involved in that scenario. In the case of dealing with urban public space, the analysis must consider a large group of citizens who carry out different activities on it. The purpose of this article is to analyze the human roles and the human acceptance when the robotic technology is introduced in the last mile distribution of goods in urban areas. In this work, we start with the description of the Human-Robot Interaction (HRI) scenario for last mile goods delivery, we describe the human roles and we propose a set of relevant indicators to evaluate the human acceptance for this task. Finally, we evaluate the human acceptance through qualitative interviews and quantitative surveys. The study has been done for the peer end-users and bystanders human roles and around 100 people participated., Work supported under the European project LogiSmile with the grant (EIT-UM-2020-22140) and by the Spanish State Research Agency through the ROCOTRANSP project with the grant (PID2019-106702RB-C21 / AEI / 10.13039/501100011033) and AUDEL project (TED2021-131759A-I00)), Peer Reviewed, Postprint (author's final draft)

Within the next years, industrial collaborative robots will collaborate with human operators, whose safety might be compromised. To ensure a safe collaboration, robots should estimate the risk of collision with respect to the pose and motion of operators within the shared workspace. A common approach is to compute and maintain a minimum distance between humans and robots during tasks' execution. Nevertheless, separation-based solutions do not capture the real dynamics of the human-robot interaction and tend to be rather conservative, avoiding a close human-robot collaboration. In this chapter, we explore the concept of time-to-contact (TTC) as a softer trigger of safety stops in collaborative scenarios where humans and robots are in constant closeness. Particularly, we propose a TTC formulation and study its advantages with respect to two approaches based on the protective distance proposed by the ISO standards. We compared the three methods in some representative cases extracted from an example of a collaborative task. The evaluation is firstly done in simulation and then in a more realistic setup with a simulated human, aiming for repeatability, and a real robot. Furthermore, we showcased our approach in a demo of a complete collaborative task. TTC allows robots to operate closer to humans and for longer times before a safety stop is issued, which benefits long-term productivity. This later stop produces shorter human-robot distances, which might affect safety. However, the increment in time that the robot moves before stopping (productivity) is far greater than the reduction in the distance (safety). Hence, we can state that TTC greatly improves productivity while slightly compromising safety. In conclusion, our work demonstrates that TTC is a smoother, but still safe, collision risk estimator for close human-robot collaboration., Peer Reviewed, Postprint (published version)

© 2023 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND, We propose a novel approach to enhance the 3D body pose estimation of a person computed from videos captured from a single wearable camera. The main technical contribution consists of leveraging high-level features linking first- and third-views in a joint embedding space. To learn such embedding space we introduce First2Third-Pose, a new paired synchronized dataset of nearly 2000 videos depicting human activities captured from both first- and third-view perspectives. We explicitly consider spatial- and motion-domain features, combined using a semi-Siamese architecture trained in a self-supervised fashion. Experimental results demonstrate that the joint multi-view embedded space learned with our dataset is useful to extract discriminatory features from arbitrary single-view egocentric videos, with no need to perform any sort of domain adaptation or knowledge of camera parameters. An extensive evalu- ation demonstrates that we achieve significant improvement in egocentric 3D body pose estimation per- formance on two unconstrained datasets, over three supervised state-of-the-art approaches. The collected dataset and pre-trained model are available for research purposes., This work has been partially supported by projects PID2020-120 049RB-I00 and PID2019-110977GA-I00 funded by MCIN/ AEI/10.13039/501100 011033 and by the “European Union NextGener-ationEU/PRTR”, as well as by grant RYC-2017-22563 funded by MCIN/ AEI /10.13039/501100 011033 and by “ESF Investing in your future”, and network RED2018-102511-T funded by MCIN/ AEI, Peer Reviewed, Postprint (published version)

The ability to install social intelligence protocols in robots in order for them to exhibit conversational skills has made them ideal tools for delivering services with a high cognitive and low emotional load. Little is known about how this capability influences the customer experience and the intention to continue receiving these services. Experiences were assessed in a study simulating customer-facing service delivery, and the constructs of the technology readiness index and stated gender were analysed as possible moderators in a quasi-experiment. Hedonic quality was the most relevant factor explaining attitude, and attitude explained intention to use as well as social influence. As for the constructs of technological readiness and gender, optimism and innovativeness seem to be the most likely candidates for moderating the other variables. The most optimistic and the most innovative route would be for the main actors to continue adapting to social robot technology in the future., This work was partially funded by the European Union’s Horizon 2020 under ERC Advanced Grant CLOTHILDE (no. 741930); by MCIN/ AEI /10.13039/501100011033 and by the ”European Union NextGenerationEU/PRTR” under the project ROB-IN (PLEC2021-007859); by the Research Council of Norway under the project SECUROPS (INT-NO/0875); and by the ”European Union NextGenerationEU/PRTR” through CSIC’s Thematic Platforms (PTI+ Neuro-Aging)., Peer Reviewed, Postprint (author's final draft)

The version of record is available online at: https://dx.doi.org/10.1007/s11257-022-09327-w, Task planning in human–robot environments tends to be particularly complex as it involves additional uncertainty introduced by the human user. Several plans, entailing few or various differences, can be obtained to solve the same given task. To choose among them, the usual least-cost plan criteria is not necessarily the best option, because here, human constraints and preferences come into play. Knowing these user preferences is very valuable to select an appropriate plan, but the preference values are usually hard to obtain. In this context, we propose the Space-of-Plans-based Suggestions (SoPS) algorithms that can provide suggestions for some planning predicates, which are used to define the state of the environment in a task planning problem where actions modify the predicates. We denote these predicates as suggestible predicates, of which user preferences are a particular case. The first algorithm is able to analyze the potential effect of the unknown predicates and provide suggestions to values for these unknown predicates that may produce better plans. The second algorithm is able to suggest changes to already known values that potentially improve the obtained reward. The proposed approach utilizes a Space of Plans Tree structure to represent a subset of the space of plans. The tree is traversed to find the predicates and the values that would most increase the reward, and output them as a suggestion to the user. Our evaluation in three preference-based assistive robotics domains shows how the proposed algorithms can improve task performance by suggesting the most effective predicate values first., This work has been partially supported by the ERC project Clothilde (ERC-2016-ADG-741930); by MCIN/AEI/10.13039/501100011033 under the project CHLOE-GRAPH (PID2020-118649RB-l00); by the European Union NextGenerationEU/PRTR under the project ROB-IN (PLEC2021-007859); and by the CHIST-ERA project COHERENT (EPSRC EP/V062506/1), Peer Reviewed, Postprint (published version)

© 2023 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works., Teaching complex manipulation skills, such as folding garments, to a bi-manual robot is a very challenging task, which is often tackled through learning from demonstration. The few datasets of garment-folding demonstrations available nowadays to the robotics research community have been either gathered from human demonstrations or generated through simulation. The former have the great difficulty of perceiving both cloth state and human action as well as transferring them to the dynamic control of the robot, while the latter require coding human motion into the simulator in open loop, i.e., without incorporating the visual feedback naturally used by people, resulting in far-from-realistic movements. In this article, we present an accurate dataset of human cloth folding demonstrations. The dataset is collected through our novel virtual reality (VR) framework, based on Unity’s 3D platform and the use of an HTC Vive Pro system. The framework is capable of simulating realistic garments while allowing users to interact with them in real time through handheld controllers. By doing so, and thanks to the immersive experience, our framework permits exploiting human visual feedback in the demonstrations while at the same time getting rid of the difficulties of capturing the state of cloth, thus simplifying data acquisition and resulting in more realistic demonstrations. We create and make public a dataset of cloth manipulation sequences, whose cloth states are semantically labeled in an automatic way by using a novel low-dimensional cloth representation that yields a very good separation between different cloth configurations., The research leading to these results receives funding from the European Research Council (ERC) from the European Union Horizon 2020 Programme under grant agreement no. 741930 (CLOTHILDE: CLOTH manIpulation Learning from DEmonstrations) and project SoftEnable (HORIZONCL4-2021-DIGITAL-EMERGING-01-101070600). Authors also received funding from project CHLOE-GRAPH (PID2020-118649RB-I00) funded by MCIN/ AEI /10.13039/501100011033 and COHERENT (PCI2020-120718-2) funded by MCIN/ AEI /10.13039/501100011033 and cofunded by the ”European Union NextGenerationEU/PRTR”., Peer Reviewed, Postprint (author's final draft)

Robotic manipulation of cloth is a complex task because of the infinite-dimensional shape-state space of textiles, which makes their state estimation very difficult. In this paper we introduce the dGLI Cloth Coordinates, a finite low-dimensional representation of cloth states that allows us to efficiently distinguish a large variety of different folded states, opening the door to efficient learning methods for cloth manipulation planning and control. Our representation is based on a directional derivative of the Gauss Linking Integral and allows us to represent spatial as well as planar folded configurations in a consistent and unified way. The proposed dGLI Cloth Coordinates are shown to be more accurate in the representation of cloth states and significantly more sensitive to changes in grasping affordances than other classic shape distance methods. Finally, we apply our representation to real images of a cloth, showing that with it we can identify the different states using a distance-based classifier., This work was developed under the project CLOTHILDE which has received funding from the European Research Council (ERC) under the EU-Horizon 2020 research and innovation programme (grant agreement No. 741930). M. Alberich-Carramiñana is also with the Barcelona Graduate School of Mathematics (BGSMath) and the Institut de Matemàtiques de la UPC-BarcelonaTech (IMTech), and she and J. Amorós are partially supported by the Spanish State Research Agency AEI/10.13039/501100011033 grant PID2019-103849GB-I00 and by the AGAUR project 2021 SGR 00603 Geometry of Manifolds and Applications, GEOMVAP. J. Borràs is supported by the Spanish State Research Agency MCIN/ AEI /10.13039/501100011033 grant PID2020-118649RB-I00 (CHLOE-GRAPH project)., Peer Reviewed, Postprint (published version)

The expected increase of hydrogen fuel cell vehicles has motivated the emergence of a significant number of studies on Hydrogen Refuelling Station (HRS). At the same time, many projects to build new HRS have been set in motion around the world. Some of the main HRS topics in the literature are sizing, location, design optimization, and optimal operation. On-site green HRS, where hydrogen is produced locally from green renewable energy sources, have received special attention due to their contribution to decarbonization. This kind of HRS are complex systems whose hydraulic and electric linked topologies may include renewable energy sources, electrolyzers, buffer hydrogen tanks, compressors and batteries, among other components. This work addresses the modelling and operation of a real on-site green HRS that is set to be built in Zaragoza, Spain, which will be the 6th in the country. This plant can produce hydrogen either from solar irradiance energy or from the utility grid and is designed for three different types of services: light-duty and heavy-duty fuel cell vehicles and gas containers. Specifically, the case of study HRS consists of six hydrogen tanks, four compressors and three electrical energy sources. To the best of the author’s knowledge, this thesis develops an unprecedented model of such a flexible HRS design and validates it through simulation by implementing a finite-state machine as the control method. Annual results show some refuelling processes being left uncompleted which hints at the necessity of a predictive control strategy. Moreover, the model presented confirms that for the case of study HRS, low-carbon hydrogen production is possible under the currently drafted European Union standards, with emission savings in the 75-96 % compared to the fossil-fuel comparator and a hydrogen emission intensity in the 3-24 gCO2eq/MJ. In the literature, there is a lack of online control solutions developed for HRS, even more in the form of optimal online cont, El previsible aumento del número de vehículos de pila de combustible de hidrógeno ha motivado la aparición de un importante número de estudios sobre las Estación de Repostaje de Hidrógeno (ERH). Al mismo tiempo, se han puesto en marcha numerosos proyectos alrededor del mundo para construir nuevas ERH. Algunos de los temas principalmente tratados en la literatura sobre ERH son el dimensionamiento, la ubicación, la optimización del diseño y el funcionamiento ópti- mo. Las ERH verdes de producción in situ, aquellas en las que el hidrógeno se produce localmente a partir de fuentes de energía renovables, han recibido especial atención por su contribución a la descarbonización. Este tipo de ERH son sistemas complejos cuyas topologías hidráulicas y eléctricas estan enlazadas y pueden incluir diversos componentes tales como fuentes de energía renovable, electrolizadores, depósitos de hidrógeno, compresores y baterías, entre otros. Este tra- bajo aborda la modelización y operación de una planta verde rea l que se construirá en Zaragoza, España, y que será la sexta del país. Esta planta puede producir hidrógeno a partir de energía de irradiancia solar o de la red eléctrica y está diseñada para tres tipos de servicios diferentes: vehículos ligeros y pesados de pila de combustible y contenedores de gas. En concreto, el caso de estudio ERH consta de seis depósitos de hidrógeno, cuatro compresores y tres fuentes de energía eléctrica. Hasta donde alcanza el conocimiento del autor, esta tesis desarrolla un modelo sin precedentes de un diseño ERH tan flexible y lo valida mediante simulación implementando una máquina de estados finitos como método de control. Los resultados anuales muestran que algunos de los procesos de repostaje quedan incompletos, lo que apunta a la necesidad de una estrategia de control predictivo. Además, el modelo presentado confirma que, para el caso de estudio, la producción de hidrógeno con bajas emisiones de carbono es posible según el actual borrador del e, El previsible augment del nombre de vehicles de pila de combustible d’hidrogen ha motivat l’a- parició d’un important nombre d’estudis sobre les Estació de Recàrrega d’Hidrogen (ERH). Al mateix temps, s’han posat en marxa nombrosos projectes al voltant del món per a construir noves ERH. Alguns dels temes principalment tractats en la literatura sobre ERH són el dimen- sionament, la ubicació, l’optimització del disseny i el funcionament òptim. Les ERH verdes de producció in situ, aquelles en les quals l’hidrogen es produeix localment a partir de fonts d’ener- gia renovables, han rebut especial atenció per la seva contribució a la descarbonització. Aquest tipus de ERH són sistemes complexos on les topologies hidràuliques i elèctriques estan enllaçades i inclouen components diversos com ara fonts d’energia renovable, electrolizadors, dipòsits d’hi- drogen, compressors i bateries, entre altres components. Aquest treball aborda la modelització i operació d’una planta d’hidrogen verd real que es construirà a Zaragoza, Espanya, i que serà la sisena del país. Aquesta planta pot produir hidrogen a partir d’energia d’irradiància solar o de la xarxa elèctrica i està dissenyada per a tres tipus de serveis diferents: vehicles lleugers i pesats de pila de combustible i contenidors de gas. En concret, el cas d’estudi ERH consta de sis dipòsits d’hidrogen, quatre compressors i tres fonts d’energia elèctrica. Fins on arriba el conei- xement de l’autor, aquesta tesi desenvolupa un model sense precedents d’un disseny ERH tan flexible i el valida mitjançant simulació implementant una màquina d’estats finits com a mètode de control. Els resultats anuals mostren que alguns dels processos de proveïment d’hidrogen queden incomplets, la qual cosa apunta a la necessitat d’una estratègia de control predictiu. A més, el model presentat confirma que, per al cas d’estudi, la producció d’hidrogen amb baixes emissions de carboni és possible segons l’actual esborrany del estándard de la Unió Europea

In this Master’s Thesis, a control architecture for a microgrid containing a Fuel Cell, Superca- pacitor and lithium-ion Battery (BT) has been developed, with the goal to be implementable in the microgrid located in the FC Lab of IRI. The mentioned microgrid has been modified with the addition of the BT, for which hardware and software had to be developed. As the topology of the microgrid is full-active, which means that every element is connected to a DC bus through a power converter, the control of this converters has been also designed, simulated and implemented. In order of effectively integrating the BT into the microgrid, a battery voltage model has been developed and fitted, and then used in the BT SoC estimation using EKF. Then, a strategy based in Filter-Based Control (FBC) has been developed and simulated, taking as case of application the behaviour of a van in the WLTP cycle. It has been found that the control could successfully manage the power distribution under this case, maintaining the bus voltage at the desired reference and the BT and SC SoCs within the desired ranges, En aquesta tesi de màster, s’ha desenvolupat l’arquitectura de control per una micro- xarxa que conté una pila de combustible (PC), supercondensadors (SC) i bateries d’ió de liti (BT), amb l’objectiu de ser implementable en la microxarxa que es troba al Laboratori de PC de l’IRI. La microxarxa mencionada s’ha modificat per afegir la BT, per la qual cosa va ser necessari el desenvolupament de hardware i software. Com la topologia de la microxarxa és completament activa, cosa que vol dir que tots els elements estan connectats al bus DC a través d’un convertidor de potència, el control d’aquests conver- tidors també s’ha hagut de dissenyar, simular i implementar. Per tal d’integrar efectivament la BT a la microxarxa, s’ha desenvolupat i ajustat un model de la tensió de bateria, per després utilitzar-lo per estimar el SoC de la BT utilitzant EKF. Després, s’ha desenvolupat i simulat una estratègia de control basada en filtres, prenent com a cas d’aplicació el comportament d’una furgoneta en el cicle WLTP. S’ha trobat que el control podia gestionar exitosament la distribució de potència per aquest cas, mantenint la tensió de bus a la referència desitjada i els SoCs de SC i BT dins dels rangs desitjats, En esta tesis de master, se ha desarrollado la arquitectura de control para una microred que contine una pila de combustible (PC), supercondensadores (SC) i baterias de ión de litio (BT), con el objetivo de ser implementable en la microred que se encuentra en el Laboratorio de PC del IRI. La microred mencionada se ha modificado para añadir la BT, por la cual cosa fue necesario el desarrollo de hardware y software. Como la topología de la microred es completamente activa, cosa que significa que todos los elementos están conectados al bus DC a través de un convertidor de potencia, el control de estos convertidores también se ha tenido que diseñar, simular e implementar. Con tal de integrar efectivamente la BT en la microred, se ha desarrollado y ajustado un modelo de la tensión de batería, par después utilizarlo para estimar el SoC de la BT utilizando EKF. Después, se ha desarrollado y simulado una estrategia de control basada en filtros, tomando como caso de aplicación el comportamiento de una furgoneta en el ciclo WLTP. Se ha encontrado que el control podía gestionar exitosamente la distribución de potencia para este caso, manteniendo la tensión de bus en la referencia deseada y los SoCs de SC y BT dentro de los rangos deseados

Fossil fuels -including coal, oil, and natural gas- have been powering economies for over 150 years, and currently supply about 80% of the world's energy. But this number is meant to decrease drastically before the end of the century to prevent global warming. Russia’s invasion of Ukraine has created shock waves in global energy markets, leading to price volatility, supply shortages, security issues and economic uncertainty, leading us to the biggest energy crisis of the history. To deal with these problems, our society has to take serious actions to optimize the electric supply in the world while taking into account intermittent energy sources such as solar, wind or hydraulic power. As a consequence, the target of climate neutrality by 2050 has encouraged the growth of renewable energy in Europe: in 2020, around one-fifth of the European electricity was generated from wind and solar electricity, surpassing fossil-based electricity generation. This same year, electricity generation from coal decreased by almost 50% since 2015, which is equivalent to avoiding around 320 Mt CO2 per year1. Smart grids have the potential to optimize the efficiency, reliability, economics, and sustainability of the production, distribution, and consumption of electrical energy. In 2021, the Smart Grid Index benchmarked a total of 86 Smart Grid utilities across 37 countries2, with Enedis achieving the number one position. Indeed, a classical electrical grid is defined as a reliable integrated power delivery system consisting of interconnected Distributed Energy Resources (DERs), which has the purpose of satisfying load demands without any interruption. However, introducing renewable energies in the grid requires some predictive measures in order to guarantee the reliability and stability of the energy supply as they are highly influenced by weather conditions, economic situations, and environmental issues. As a solution, Smart Grids can handle the presence of uncertainties as well as ensu

© 2023 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works., In this work, we present a method for Multi-Object Tracking (MOT) that uses unsynchronized multimodal detections from a configurable set of sensors such as cameras, radars and lidars. All the information is processed from the sensors with modality-specific detectors and then combined in the MOT module that incorporates a Kalman filter and tracklet management logic. To be able to deploy our system in real-world driving applications, we handle localization errors, misclassifications and partial bounding-box detections from the object detector in the MOT module. We show promising results and compare them with respect to competing approaches in two challenging real-world scenarios, including a traffic jam chauffeur as well as a traffic monitoring application on highways., This work has been supported by the project MoHuCo PID2020-120049RB-I00 funded by MCIN/AEI/10.13039/501100011033 and by the Government of Catalonia under 2020 DI 105., Peer Reviewed, Postprint (author's final draft)

© 2023 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.”, Photometric stereo is the problem of jointly inferring the 3D reconstruction, reflectance, lighting and specularities of an object from a set of visual signals. Recently, some variational, uncalibrated, unsupervised and unified formulations have provided robust solutions to the problem while reducing the prior knowledge about the shape geometry or the lighting conditions. Unfortunately, these approaches cannot still produce solutions with an ample variety of details in the shape. That is mainly due to the non-convex and non-linear nature of the problem which requires the best initialization as possible. In this context, we propose a fully interpretable formulation that combines a physically-aware image formation model under perspective projection with a minimal detail-aware initialization and that it can handle general lighting. As a result, our formulation can consider multiple scenarios composed of unknown complex geometries and lighting patterns. Experimental results on challenging synthetic and real datasets show the effectiveness of our approach to capture more fine details, outperforming state-of-the-art techniques in terms of 3D reconstruction., This work has been supported by the project MoHuCo PID2020-120049RB-I00 funded by MCIN/ AEI /10.13039/501100011033, Peer Reviewed, Postprint (author's final draft)

© 2024 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works., Solar energy adoption is moving at a rapid pace. The variability in solar energy production causes grid stability issues and hinders mass adoption. To solve these issues, more accurate photovoltaic power forecasting systems are needed. In intra-hour forecasting, the most challenging issue is high output fluctuations due to cloud motion, which can occlude the sun. Using ground-based sky images, this paper proposes two convolutional neural network models for intra-hour nowcasting and forecasting that incorporate physical information on sun motion and cloud coverage by means of the sun area mean pixel intensity. Particularly, our models exploit that information instead of relying exclusively on photovoltaic output history data as it is standard in state of the art. Taking advantage of sun position and cloud coverage information, we were able to reduce the overall root mean squared error for the nowcasting task, making the model more accurate especially during cloudy days, and obtaining competitive results on forecasting. Moreover, our models are more robust against artifacts such as occlusion and noisy observations., This work has been supported by the project MoHuCo PID2020-120049RB-I00 funded by MCIN/ AEI /10.13039/501100011033., Peer Reviewed, Postprint (author's final draft)

Reinforcement Learning (RL) of trajectory data has been used in several fields, and it is of relevance in robot motion learning, in which sampled trajectories are run and their outcome is evaluated with a reward value. The responsibility on the performance of a task can be associated to the trajectory as a whole, or distributed throughout its points (timesteps). In this work, we present a novel method for attributing the responsibility of the rewards to each time-step separately by using Mutual Information (MI) to bias the model fitting of a trajectory., Postprint (author's final draft)

Unmanned Aerial Vehicles (UAVs) have become an interesting research topic in the robotics field. Their maneuverability, agility and capacity to navigate through un- orthodox environments has gained the attention from both industrial and scientific communities. Their range of applications vary from inspection in hazardous envi- ronments to search and rescue tasks. However, most state-of-the-art control schemes for UAVs assume a simple dynamic model. In many cases, they neglect the effect of the dynamics of the actuators when controlling the platform and, those which consider them, usually require a formulation to constraint the state of the vehicle, which is not available for some optimal solvers. In this work, we present a method to consider the actuator dynamics of a UAV when formulating an Optimal Control Problem (OCP). In particular, we analyze the benefits of considering the actua- tor dynamics within a higher-end Model Predictive Control (MPC) architecture, with the aim of improving the controllability of these platforms for dynamic and aggressive maneuvers. We’ve found interest in this kind of challenging trajectories because they require the exploitation of the full potential of the MPC approach due to the underactuated nature of the UAVs. In order to successfully achieve it, we need an algorithm capable of solving a non-linear OCP in very short time, hence we have chosen the usage of Differential Dynamic Programming (DDP). Building upon Crocoddyl, an open source implementation based on DDP, we have incorporated the required changes to embed the dynamics of the actuators in the original for- mulation. The resulting optimal solver has been validated through simulated case studies, comparing its performance with classical approaches.

Extended Abstract presentat al workshop de Prediction and Anticipation Reasoning in Human Robot Interaction (https://www.iri.upc.edu/workshops/pred-ant-hri/cfp.html), a la conferència ICRA 2022. Aquest article descriu un treball en progrés que es preveu afegir a la tesi. A dia d'aquesta descripció (15/07/2022), encara no s'ha publicat un enllaç al document presentat al workshop., We introduce INtuitive PROgramming 2 (INPRO2), an improvement over our previous INPRO framework for learning board exercises via demonstrations. INPRO2 makes use of our Online Action Recognition through Unification (OARU) algorithm, which maintains and extends as needed a library of STRIPS action schemata that represent the dynamics, rules and goal of the exercise. OARU operates on a sequence of states shown by the user. Each state transition is either used to learn a new action, or is recognized as an instance of one action currently present in the library, possibly refining it. We have extended OARU to support negative examples (i.e. invalid moves that show forbidden state transitions) in order to increase the complexity of the exercises that can be learned. This new OARU's feature is exploited through another crucial element of INPRO2: its ability to proactively ask for the legality of certain moves to the user in critical situations, and fix overly permissive actions. We show an example of a typical INPRO2 learning session. We also outline a plan for a user study that will serve to assess the proactive behavior of the robot., Preprint

Comencem amb una recopilació dels cervells artificials que han aparegut en obres de ficció, ja sigui incorporats en robots o bé com intel·ligències col·lectives i consciències d’altres tipus, tant de substrat tecnològic com biològic, així com els cervells híbrids dels ciborgs. Després fem un breu repàs de l’estat de l’art tant pel que fa als implants cerebrals, invasius i no-invasius, com als robots cognitius i els reptes que plantegen, tan diferents dels dels robots industrials. Finalment, descrivim algunes sèries, pel·lícules i narratives de ciència-ficció que ofereixen especulacions distòpiques d’on podria conduir-nos la interacció cada vegada més estreta amb la tecnologia, així com propostes d’utilitzar aquestes ficcions com una eina pedagògicament engrescadora tant per a estudiants amb perfil tecnocientífic com per als docents i també per a qualsevol persona aficionada a reflexionar sobre alternatives de futur. La tendència sembla ser que la frontera entre cervells naturals i artificials anirà sent cada vegada més difusa i el més probable és que, en comptes de produir-se una singularitat, ens encaminem cap a la proliferació de cervells híbrids. El desig seria que les millors prestacions de les màquines es combinin amb les millors qualitats humanes, i el major risc vindria de les fractures i discriminacions que es podrien produir., Postprint (author's final draft)

This paper presents a model-free methodology for the placement of pressure sensors in water distribution networks (WDNs) with the aim of performing leak detection/localization tasks. The approach is based on a custom genetic algorithm (GA) optimization scheme, which considers a population whose individuals are binary vectors encoding the network nodes with/without sensors. The optimization process pursues the minimization of a distance-based metric, computed considering the pipe distance from the possible sensors to the complete set of nodes of the network, hence removing the necessity of a hydraulic model of the WDN. The methodology is completed by means of an iterative clustering technique that seeks the enhancement of incoming individuals. The proposed methodology is tested over a well-known case study, L-TOWN from the BattLeDIM2020 challenge, in order to assess its performance., Peer Reviewed, Postprint (published version)

Being able to teach complex capabilities, such as folding garments, to a bi-manual robot is a very challenging task, which is often tackled using learning from demonstration datasets. The few garment folding datasets available nowadays to the robotics research community are either gathered from human demonstrations or generated through simulation. The former have the huge problem of perceiving human action and transferring it to the dynamic control of the robot, while the latter requires coding human motion into the simulator in open loop, resulting in far-from-realistic movements. In this article, we present a reduced but very accurate dataset of human cloth folding demonstrations. The dataset is collected through a novel virtual reality (VR) framework we propose, based on Unity’s 3D platform and the use of a HTC Vive Pro system. The framework is capable of simulating very realistic garments while allowing users to interact with them, in real time, through handheld controllers. By doing so, and thanks to the immersive experience, our framework gets rid of the gap between the human and robot perception-action loop, while simplifying data capture and resulting in more realistic sample, This work was developed in the context of the project CLOTHILDE (”CLOTH manIpulation Learning from DEmonstrations”) which has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation program (grant agreement No. 741930) and is also supported by the BURG project PCI2019-103447 funded by MCIN/ AEI /10.13039/501100011033 and by the ”European Union”., Peer Reviewed, Postprint (published version)

© 2022 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes,creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works., In this work we present a novel attention deep learning model that uses context and human intention for 3D human body motion prediction in handover human-robot tasks. This model uses a multi-head attention architecture which incorporates as inputs the human motion, the robot end effector and the position of the obstacles. The outputs of the model are the predicted motion of the human body and the predicted human intention. We use this model to analyze a handover collaborative task with a robot where the robot is able to predict the future motion of the human and use this information in it’s planner. Several experiments are performed where human volunteers fill a standard poll to rate different features, taking into account when the robot uses the prediction versus when the robot doesn’t use the prediction., Work supported under the Spanish State Research Agency through the Maria de Maeztu Seal of Excellence to IRI (MDM-2016- 0656), the ROCOTRANSP project (PID2019-106702RB-C21 / AEI / 10.13039/501100011033)) and the EU project CANOPIES (H2020- ICT2020-2-101016906)., Peer Reviewed, Postprint (author's final draft)

© 2021 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes,creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works., Recently, transfer subspace learning based approaches have shown to be a valid alternative to unsupervised subspace clustering and temporal data clustering for human motion segmentation (HMS). These approaches leverage prior knowledge from a source domain to improve clustering performance on a target domain, and currently they represent the state of the art in HMS. Bucking this trend, in this paper, we propose a novel unsupervised model that learns a representation of the data and digs clustering information from the data itself. Our model is reminiscent of temporal subspace clustering, but presents two critical differences. First, we learn an auxiliary data matrix that can deviate from the initial data, hence confers more degrees of freedom to the coding matrix. Second, we introduce a regularization term for this auxiliary data matrix that preserves the local geometrical structure present in the high-dimensional space. The proposed model is efficiently optimized by using an original Alternating Direction Method of Multipliers (ADMM) formulation allowing to learn jointly the auxiliary data representation, a nonnegative dictionary and a coding matrix. Experimental results on four benchmark datasets for HMS demonstrate that our approach achieves significantly better clustering performance then state-of-the-art methods, including both unsupervised and more recent semi-supervised transfer learning approaches., Work partially funded by projects MINECO/ERDF RyC, PID2019-110977GA-I00, JAEINT19-EX-0014, RED2018- 102511-T., Peer Reviewed, Postprint (author's final draft)

Differential dynamic programming (DDP) is a direct single shooting method for trajectory optimization. Its efficiency derives from the exploitation of temporal structure (inherent to optimal control problems) and explicit roll-out/integration of the system dynamics. However, it suffers from numerical instability and, when compared to direct multiple shooting methods, it has limited initialization options (allows initialization of controls, but not of states) and lacks proper handling of control constraints. In this work, we tackle these issues with a feasibility-driven approach that regulates the dynamic feasibility during the numerical optimization and ensures control limits. Our feasibility search emulates the numerical resolution of a direct multiple shooting problem with only dynamics constraints. We show that our approach (named Box-FDDP) has better numerical convergence than Box-DDP+ (a single shooting method), and that its convergence rate and runtime performance are competitive with state-of-the-art direct transcription formulations solved using the interior point and active set algorithms available in Knitro. We further show that Box-FDDP decreases the dynamic feasibility error monotonically—as in state-ofthe-art nonlinear programming algorithms. We demonstrate the benefits of our approach by generating complex and athletic motions for quadruped and humanoid robots. Finally, we highlight that Box-FDDP is suitable for model predictive control in legged robots., Peer Reviewed, Postprint (published version)

© 2022 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting /republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works, The human state is defined in terms of capacity, knowledge and motivation. The system has been implemented in a standardised environment using the Planning Domain Definition Language (PDDL) and the modular ROSPlan framework, and we have validated the approach in multiple simulation settings. Our results show that using the human model fosters an appropriate task allocation while allowing failure anticipation, replanning in time to prevent it., Peer Reviewed, Postprint (author's final draft)

© 2022 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting /republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works, Benchmarking of robotic manipulations is one of the open issues in robotic research. An important factor that has enabled progress in this area in the last decade is the existence of common object sets that have been shared among different research groups. However, the existing object sets are very limited when it comes to cloth-like objects that have unique particularities and challenges. This paper is a first step towards the design of a cloth object set to be distributed among research groups from the robotics cloth manipulation community. We present a set of household cloth objects and related tasks that serve to expose the challenges related to gathering such an object set and propose a roadmap to the design of common benchmarks in cloth manipulation tasks, with the intention to set the grounds for a future debate in the community that will be necessary to foster benchmarking for the manipulation of cloth-like objects. Some RGB-D and object scans are collected as examples for the objects in relevant configurations and shared in http://www.iri.upc.edu/groups/perception/ClothObjectSet, Peer Reviewed, Postprint (author's final draft)

© 2022 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes,creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works., Simultaneous camera pose, 4D reconstruction of an object and deformation clustering from incomplete 2D point tracks in a video is a challenging problem. To solve it, in this work we introduce a union of piecewise subspaces to encode the 4D shape, where two modalities based on B-splines and Catmull-Rom curves are considered. We demonstrate that formulating the problem in terms of B-spline or Catmull-Rom functions, allows for a better physical interpretation of the resulting priors while C1 and C2 continuities are automatically imposed without needing any additional constraint. An optimization framework is proposed to sort out the problem in a unified, accurate, unsupervised and efficient manner. We extensively validate our claims on a wide range of human motions, including articulated and continuous deformations as well as those cases with noisy and missing measurements where our approach provides competing joint solutions., Peer Reviewed, Postprint (author's final draft)

The social ability of humans to provide active feedback during conversations is known as backchannelling. Recent work has recognised the importance of endowing robots with such social behaviour to make interactions more natural. Nonetheless, very little is known about how backchannelling should be designed in order to be detected and whether it can have an impact on users' behaviour and performance in cooperative tasks. In this article, we aim at evaluating the legibility of robot's backchannelling behaviour on Persons with Dementia (PwDs) and its effect on their performance when playing cognitive training exercises. Aiming to do so, a TIAGo robot was endowed with backchannelling behaviour generated by combining verbal and non-verbal cues. To evaluate our system, two user studies were carried out, in which the social signal was provided first by a human therapist and later on by a robot. Results indicate that patients were capable of identifying such kind of feedback. Nonetheless, our findings pointed out a significant difference in terms of performance between the two studies. They reveal how patients in the study with the robot overused the feedback to obtain the correct answer, putting in place a cheating mechanism that has led them to significantly worsen their performance. We conclude our work by discussing the implications of our findings when deploying robots in sensitive roles and possible solutions to address such unexpected behaviours., Peer Reviewed, Postprint (author's final draft)

Socially assistive robots have the potential to augment and enhance therapist’s effectiveness in repetitive tasks such as cognitive therapies. However, their contribution has generally been limited as domain experts have not been fully involved in the entire pipeline of the design process as well as in the automatisation of the robots’ behaviour. In this article, we present aCtive leARning agEnt aSsiStive bEhaviouR (CARESSER), a novel framework that actively learns robotic assistive behaviour by leveraging the therapist’s expertise (knowledge-driven approach) and their demonstrations (data-driven approach). By exploiting that hybrid approach, the presented method enables in situ fast learning, in a fully autonomous fashion, of personalised patient-specific policies. With the purpose of evaluating our framework, we conducted two user studies in a daily care centre in which older adults affected by mild dementia and mild cognitive impairment (N = 22) were requested to solve cognitive exercises with the support of a therapist and later on of a robot endowed with CARESSER. Results showed that: (i) the robot managed to keep the patients’ performance stable during the sessions even more so than the therapist; (ii) the assistance offered by the robot during the sessions eventually matched the therapist’s preferences. We conclude that CARESSER, with its stakeholder-centric design, can pave the way to new AI approaches that learn by leveraging human–human interactions along with human expertise, which has the benefits of speeding up the learning process, eliminating the need for the design of complex reward functions, and finally avoiding undesired states., Peer Reviewed, Postprint (published version)

© 2022 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting /republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works, One of the major challenges in learning from demonstration is to teach the robot a wider set of task features than the plain trajectories to be followed. In this sense, one key parameter is stiffness, i.e., the rigidity that the manipulator should exhibit when performing a task. The required robot stiffness is often not known a priori and varies along the execution of the task, thus its profile needs to be inferred from the demonstrations. In this work, we propose a novel, force-based algorithm for inferring time-varying stiffness profiles, leveraging the relationship between stiffness and tracking error, and involving human-robot interaction. We begin by gathering a set of demonstrations with kinesthetic teaching. Then, the robot executes a perturbed reference, obtained from these demonstrations by means of Gaussian process regression, and the human intervenes if the perturbation makes the manipulator deviate from its expected behaviour. Human intervention is measured and used to infer the desired control stiffness. In the experiments section, we show that our algorithm can be combined with different types of force sensors, and provide suitable processing algorithms. Our approach correctly infers the stiffness profiles from the force and electromyography sensors, their combination permitting also to comply with the physical constraints imposed by the environment. This is demonstrated in three experiments of increasing complexity: a motion in free Cartesian space, a rigid assembly task, and bed-making., Peer Reviewed, Postprint (author's final draft)