This work considers alternative strategies to mainstream digital approaches to signal processing - namely analog and neuromorphic solutions, for increased computing efficiency. In the context of a speech recognizer application, we use low-power analog approaches for the signal conditioning and basic auditory feature extraction, while using a neuromorphic IC for building a dendritic classifier that can be used as a low-power word spotter. In doing so, this work also aspires to posit the significance of dendrites in neural computation.

We present a novel control strategy for running which is robust to disturbances, and makes excellent use of passive dynamics for energy economy. The motivation for our control strategy is based on observations of animals, which are able to economically walk and run over varying terrain and ground dynamics. It is well-known that steady-state animal running can be approximated by spring-mass models, but these passive dynamic models describe only steady-state running and are sensitive to disturbances that animals can accommodate. While animals rely on their passive dynamics for energy economy, they also incorporate active control for disturbance rejection. The same approach can be used for spring-mass walking and running, but an active controller is needed that interferes minimally with the passive dynamics of the system. We demonstrate, in simulation, how force control combined with a leg spring stiffness tuned for the desired hopping frequency provides robustness to disturbances on a model for robot hopping, while maintaining the energy economy of a completely passive system during steady-state operation. Our strategy is promising for robotics applications, because there is a clear distinction between the passive dynamic behavior of the model and the active controller, it does not require sensing of the environment, and it is based on a sound theoretical background that is compatible with existing high-level controllers for ideal spring-mass models. / Graduation date: 2012

abstract: Swarms of low-cost, autonomous robots can potentially be used to collectively perform tasks over large domains and long time scales. The design of decentralized, scalable swarm control strategies will enable the development of robotic systems that can execute such tasks with a high degree of parallelism and redundancy, enabling effective operation even in the presence of unknown environmental factors and individual robot failures. Social insect colonies provide a rich source of inspiration for these types of control approaches, since they can perform complex collective tasks under a range of conditions. To validate swarm robotic control strategies, experimental testbeds with large numbers of robots are required; however, existing low-cost robots are specialized and can lack the necessary sensing, navigation, control, and manipulation capabilities. To address these challenges, this thesis presents a formal approach to designing biologically-inspired swarm control strategies for spatially-confined coverage and payload transport tasks, as well as a novel low-cost, customizable robotic platform for testing swarm control approaches. Stochastic control strategies are developed that provably allocate a swarm of robots around the boundaries of multiple regions of interest or payloads to be transported. These strategies account for spatially-dependent effects on the robots' physical distribution and are largely robust to environmental variations. In addition, a control approach based on reinforcement learning is presented for collective payload towing that accommodates robots with heterogeneous maximum speeds. For both types of collective transport tasks, rigorous approaches are developed to identify and translate observed group retrieval behaviors in Novomessor cockerelli ants to swarm robotic control strategies. These strategies can replicate features of ant transport and inherit its properties of robustness to different environments and to varying team compositions. The approaches incorporate dynamical models of the swarm that are amenable to analysis and control techniques, and therefore provide theoretical guarantees on the system's performance. Implementation of these strategies on robotic swarms offers a way for biologists to test hypotheses about the individual-level mechanisms that drive collective behaviors. Finally, this thesis describes Pheeno, a new swarm robotic platform with a three degree-of-freedom manipulator arm, and describes its use in validating a variety of swarm control strategies. / Dissertation/Thesis / Doctoral Dissertation Mechanical Engineering 2017

abstract: Learning from the anatomy of leaves, a new approach to bio-inspired passive evaporative cooling is presented that utilizes the temperature-responsive properties of PNIPAm hydrogels. Specifically, an experimental evaporation rate from the polymer, PNIPAm, is determined within an environmental chamber, which is programmed to simulate temperature and humidity conditions common in Phoenix, Arizona in the summer. This evaporation rate is then used to determine the theoretical heat transfer through a layer of PNIPAm that is attached to an exterior wall of a building within a ventilated cavity in Phoenix. The evaporation of water to the air gap from the polymer layer absorbs heat that could otherwise be conducted to the interior space of the building and then dispels it as a vapor away from the building. The results indicate that the addition of the PNIPAm layer removes all heat radiated from the exterior cladding, indicating that it could significantly reduce the demand for air conditioning at the interior side of the wall to which it is attached. / Dissertation/Thesis / Masters Thesis Built Environment 2018

Orientadores: Romis Ribeiro de Faissol Attux, Ricardo Suyama / Tese (doutorado) - Universidade Estadual de Campinas, Faculdade de Engenharia Elétrica e de Computação / Made available in DSpace on 2018-08-19T01:01:27Z (GMT). No. of bitstreams: 1 Nadalin_EvertonZaccaria_D.pdf: 2615054 bytes, checksum: 5d288f06df05b7075bf283243319df70 (MD5) Previous issue date: 2011 / Resumo: Neste trabalho, foi estudado o problema de Separação Cega de Fontes (BSS), com ênfase nos casos chamados de subparametrizados, isto é, em que o número de fontes é maior do que o de misturas. A primeira contribuição proposta foi a de um limitante relacionado ao erro de inversão intrínseco ao problema quando é utilizada uma estrutura linear de separação. As outras contribuições estão relacionadas à hipótese de que as fontes são esparsas: i) uma proposta de metodologia híbrida, que se utiliza de conceitos baseados em independência e esparsidade dos sinais de forma simultânea para estimar tanto o sistema misturador quanto o número de fontes existentes em misturas com dois sensores; ii) a utilização de ferramentas de otimização baseadas na operação do sistema imunológico para a estimação do sistema misturador em problemas intrinsecamente multimodais; por fim, iii) uma proposta de utilização de um critério baseado em esparsidade para separação de fontes, sendo derivado um processo de otimização baseado na norma ?1 para este fim / Abstract: In this work, we studied the problem of Blind Source Separation (BSS), with emphasis on cases referred to as underdetermined, which occur when the number of sources is greater than the number of mixtures. The first contribution was a proposal of a bound to the inversion error that is intrinsic to the problem when a linear structure is used to perform separation. The other contributions are related to the hypothesis that the signals of the sources are sparse: i) the proposal of a hybrid methodology that employs concepts based on signal independence and sparsity to simultaneously estimate both the mixing system and the number of existing sources in mixtures with two sensors; ii) the use of optimization tools based on the modus operandi of the immune system to estimate the mixing system in problems that are inherently multimodal; finally, iii) the use of a criterion based on sparsity for source separation, which is derived from an optimization process based on the ?1 norm / Doutorado / Engenharia de Computação / Doutor em Engenharia Elétrica

Cette thèse s'intéresse aux mécanismes permettant de faciliter l'acquisition autonome de comportements chez les êtres vivants et propose d'utiliser ces mécanismes dans le cadre de tâches robotiques. Des réseaux de neurones artificiels sont utilisés pour modéliser certaines structures cérébrales, à la fois afin de mieux comprendre le fonctionnement de ces structures dans le cerveau des mammifères et pour obtenir des algorithmes robustes et adaptatifs de contrôle en robotique.Les travaux présentés se basent sur un modèle de l'hippocampe permettant d'apprendre des relations temporelles entre des événements perceptifs. Les neurones qui forment le substrat de cet apprentissage, appelés cellules de transition, permettent de faire des prédictions sur les événements futurs que le robot pourrait rencontrer. Ces transitions servent de support à la construction d'une carte cognitive, située dans le cortex préfrontal et/ou pariétal. Cette carte peut être apprise lors de l'exploration d'un environnement inconnu par un robot mobile et ensuite utilisée pour planifier des chemins lui permettant de rejoindre un ou plusieurs buts.Outre leur utilisation pour la construction d'une carte cognitive, les cellules de transition servent de base à la conception d'un modèle d'apprentissage par renforcement. Une implémentation neuronale de l'algorithme de Q-learning, utilisant les transitions, est réalisée de manière biologiquement plausible en s'inspirant des ganglions de la base. Cette architecture fournit une stratégie de navigation alternative à la planification par carte cognitive, avec un apprentissage plus lent, et correspondant à une stratégie automatique de bas-niveau. Des expériences où les deux stratégies sont utilisées en coopération sont réalisées et des lésions du cortex préfrontal et des ganglions de la base permettent de reproduire des résultats expérimentaux obtenus chez les rats.Les cellules de transition peuvent apprendre des relations temporelles précises permettant de prédire l'instant où devrait survenir un événement. Dans un modèle des interactions entre l'hippocampe et le cortex préfrontal, nous montrons comment ces prédictions peuvent expliquer certains enregistrements in-vivo dans ces structures cérébrales, notamment lorsqu'un rat réalise une tâche durant laquelle il doit rester immobile pendant 2 secondes sur un lieu but pour obtenir une récompense. L'apprentissage des informations temporelles provenant de l'environnement et du comportement permet de détecter des régularités. A l'opposé, l'absence d'un événement prédit peut signifier un échec du comportement du robot, qui peut être détecté et utilisé pour adapter son comportement en conséquence. Un système de détection de l'échec est alors développé, tirant parti des prédictions temporelles fournies par l'hippocampe et des interactions entre les aspects de modulation comportementale du cortex préfrontal et d'apprentissage par renforcement dans les ganglions de la base. Plusieurs expériences robotiques sont conduites dans lesquelles ce signal est utilisé pour moduler le comportement d'un robot, dans un premier temps de manière immédiate, afin de mettre fin aux actions du robot qui le mènent à un échec et envisager d'autres stratégies. Ce signal est ensuite utilisé de manière plus permanente pour moduler l'apprentissage des associations menant à la sélection d'une action, afin que les échecs répétés d'une action dans un contexte particulier fassent oublier cette association.Finalement, après avoir utilisé le modèle dans le cadre de la navigation, nous montrons ses capacités de généralisation en l'utilisant pour le contrôle d'un bras robotique. Ces travaux constituent une étape importante pour l'obtention d'un modèle unifié et générique permettant le contrôle de plates-formes robotiques variés et pouvant apprendre à résoudre des tâches de natures différentes. / This thesis takes interest in the mechanisms facilitating the autonomous acquisition of behaviors in animals and proposes to use these mechanisms in the frame of robotic tasks. Artificialneural networks are used to model cerebral structures, both to understand how these structureswork and to design robust and adaptive algorithms for robot control.The work presented here is based on a model of the hippocampus capable of learning thetemporal relationship between perceptive events. The neurons performing this learning, calledtransition cells, can predict which future events the robot could encounter. These transitionssupport the building of a cognitive map, located in the prefrontal and/or parietal cortex. The mapcan be learned by a mobile robot exploring an unknown environment and then be used to planpaths in order to reach one or several goals.Apart from their use in building a cognitive map, transition cells are also the basis for thedesign of a model of reinforcement learning. A biologically plausible neural implementation ofthe Q-learning algorithm, using transitions, is made by taking inspiration from the basal ganglia.This architecture provides an alternative strategy to the cognitive map planning strategy. Thereinforcement learning strategy requires a longer learning period but corresponds more to an automatic low-level behavior. Experiments are carried out with both strategies used in cooperationand lesions of the prefrontal cortex and basal ganglia allow to reproduce experimental resultsobtained with rats.Transition cells can learn temporally precise relations predicting the exact timing when anevent should be perceived. In a model of interactions between the hippocampus and prefrontalcortex, we show how these predictions can explain in-vivo recordings in these cerebral structures, in particular when rat is carrying out a task during which it must remain stationary for 2seconds on a goal location to obtain a reward. The learning of temporal information about theenvironment and the behavior of the robot allows the system to detect regularity. On the contrary, the absence of a predicted event can signal a failure in the behavior of the robot, whichcan be detected and acted upon in order to modulate the failing behavior. Consequently, a failure detection system is developed, taking advantage of the temporal predictions provided by thehippocampus and the interaction between behavior modulation functions in the prefrontal cortexand reinforcement learning in the basal ganglia. Several robotic experiments are conducted, inwhich the failure signal is used to modulate, immediately at first, the behavior of the robot inorder to stop selecting actions which lead to failures and explore other strategies. The signal isthen used in a more lasting way by modulating the learning of the associations leading to theselection of an action so that the repeted failures of an action in a particular context lead to thesuppression of this association.Finally, after having used the model in the frame of navigation, we demonstrate its generalization capabilities by using it to control a robotic arm in a trajectory planning task. This workconstitutes an important step towards obtaining a generic and unified model allowing the controlof various robotic setups and the learning of tasks of different natures.

A Clusterização de dados em grupos oferece uma maneira de entender e extrair informações relevantes de grandes conjuntos de dados. A abordagem em relação a aspectos como a representação dos dados e medida de similaridade entre clusters, e a necessidade de ajuste de parâmetros iniciais são as principais diferenças entre os algoritmos de clusterização, influenciando na qualidade da divisão dos clusters. O uso cada vez mais comum de grandes conjuntos de dados aliado à possibilidade de melhoria das técnicas já existentes tornam a clusterização de dados uma área de pesquisa que permite inovações em diferentes campos. Nesse trabalho é feita uma revisão dos métodos de clusterização já existentes, e é descrito um novo método de clusterização de dados baseado na identificação de comunidades em redes complexas e modelos computacionais inspirados biologicamente. A técnica de clusterização proposta é composta por duas etapas: formação da rede usando os dados de entrada; e particionamento dessa rede para obtenção dos clusters. Nessa última etapa, a técnica de otimização por nuvens de partículas é utilizada a fim de identificar os clusters na rede, resultando em um algoritmo de clusterização hierárquico divisivo. Resultados experimentais revelaram como características do método proposto a capacidade de detecção de clusters de formas arbitrárias e a representação de clusters com diferentes níveis de refinamento. / DAta clustering is an important technique to understand and to extract relevant information in large datasets. Data representation and similarity measure adopted, and the need to adjust initial parameters, are the main differences among clustering algorithms, interfering on clusters quality. The crescent use of large datasets and the possibility to improve existing techniques make data clustering a research area that allows innovation in different fields. In this work is made a review of existing data clustering methods, and it is proposed a new data clustering technique based on community dectection on complex networks and bioinspired models. The proposed technique is composed by two steps: network formation to represent input data; and network partitioning to identify clusters. In the last step, particle swarm optimization technique is used to detect clusters, resulting in an hierarchical clustering algorithm. Experimental results reveal two main features of the algorithm: the ability to detect clusters in arbitrary shapes and the ability to generate clusters with different refinement degrees

Construction is a complex activity that requires the cooperation of multiple workers. Every year, construction activities cause injuries and casualties. To make construction safer, new solutions could be provided by robotics. Robots could be employed not only to replace human workers, but also to make construction in harsh environments safe and cost-effective, paving the way for enhanced underwater infrastructure, deeper underground mining, and planetary colonization. In this thesis, we focus on the topic of collective construction, which involves the cooperation of multiple robots, by presenting a collective robot construction method of our own. Collective construction can be a more viable option than employing individual, complex robots, by potentially allowing the effective realization of large structures, while offering resilience through redundancy, analogous to insect colonies. Our approach offers a novel solution in the design trade-off between choosing the number of robots involved vs. the complexity of the robots involved. On the one hand, capable and complex robots are expensive, limiting the cost effectiveness of realizing large swarms which provide redundancy and increase the system’s resilience to faults. On the other hand, simple and inexpensive robots can be manufactured in large numbers and offer high redundancy, at the cost of limited individual capa bilities and lower performance. We use two types of robots: intelligent scaffolding and worker robots. The intelligent scaffolding acts as regular scaffolding, allowing the worker robots to navigate the structure they assemble, while also guiding and monitoring the construction of the structure. The worker robots move and connect scaffolding and building material while only knowing the local commands necessary to complete their task. This approach is loosely inspired by termite mounds, in which termites use the process of stigmergy in which they mark construction pellets with pheromones to affect the progress of construction, while navigating the struc ture that they build. Thanks to intelligent scaffolding, construction robots have a simple design that allows minimalist onboard computation and communication equipment. In this thesis, we produced a minimum viable prototype demonstrating this concept. Intelligent scaffolding is realized through smart blocks that can be laid and connected to each other. The smart blocks are capable of simple computation and communication once laid. The construction robot uses local navigation methods by line-following across the scaffolding and building blocks of the system. The blocks and construction robot both have a modular design, simplifying the process of manufacturing and repairs while maintaining a low cost. The robot and blocks use magnets to increase the margin of error during block manipulation and allow for the assembly and removal of scaffolding as well as its reuse between build sites. To communicate with the robot, the intelligent scaffolding blocks send local IR signals, similar to TV remote signals, when the robot is on top of them, minimizing the risk of global interference and keeping the system portable. To monitor the connectivity of the system throughout the life cycle of the structure, electrical connections run through each of the blocks, which indicate the status of the structure and can be used to diagnose the location of breaks in the structure for maintenance.

Les jonctions tunnel magnétiques sont des candidats prometteurs for le calcul. Mais quand elles sont réduites à des dimensions nanométriques, conserver leur stabilité devient difficile. Les jonctions tunnel magnétiques instables subissent des renversements aléatoires de leur aimantation et se comportent comme des oscillateurs stochastiques. Pourtant, la nature stochastique de ces jonctions tunnel superparamagnétiques n’est pas une faille mais un atout qui peut être utilisé pour le calcul bio-inspiré. En effet, notre cerveau a évolué de sorte qu’il puisse fonctionner dans un environnement bruité et avec des composants instables. Dans cette thèse, nous montrons plusieurs applications possibles des jonctions tunnel superparamagnétiques.Nous démontrons qu’une junction tunnel superparamagnétique peut être synchronisée en fréquence et en phase à une faible tension oscillante. De manière contre intuitive, notre expérience montre que cela peut être fait grâce à l’injection de bruit dans le système. Nous développons un modèle théorique pour comprendre ce phénomène et prédire qu’il permet un gain énergétique d’un facteur cent par rapport à la synchronisation d’oscillateurs à transfert de spin traditionnels. De plus, nous utilisons notre modèle pour étudier la synchronisation de plusieurs jonctions couplées. De nombreuses méthodes théoriques réalisant des tâches cognitives telles que la reconnaissance de motifs et la classification grâce à la synchronisation d’oscillateurs ont été proposés. Utiliser la synchronisation induite par le bruit de jonctions tunnel superparamagnétiques permettrait de réaliser ces tâches à basse énergie.Nous faisons une analogie entre les jonctions tunnel superparamagnétiques et les neurones sensoriels qui émettent des pics de tension séparés par des intervalles aléatoires. En poursuivant cette analogie, nous démontrons que des populations de jonctions tunnel superparamagnétiques peuvent représenter des distributions de probabilité et réaliser de l’inférence Bayésienne. De plus, nous démontrons que des populations interconnectées peuvent faire du calcul, notamment de l’apprentissage, des transformations de coordonnées et de la fusion sensorielles. Un tel système est faisable de manière réaliste et pourrait permettre de fabriquer des capteurs intelligents à bas coût énergétique. / Magnetic tunnel junctions are promising candidates for computing applications. But when they are reduced to nanoscale dimensions, maintaining their stability becomes an issue. Unstable magnetic tunnel junctions undergo random switches of the magnetization between their two stable states and thus behave as stochastic oscillators. However, the stochastic nature of these superparamagnetic tunnel junctions is not a liability but an asset which can be used for the implementation of bio-inspired computing schemes. Indeed, our brain has evolved to function in a noisy environment and with unstable components. In this thesis, we show several possible applications of superparamagnetic tunnel junctions.We demonstrate how a superparamagnetic tunnel junction can be frequency and phase-locked to a weak oscillating voltage. Counterintuitively, our experiment shows that this is achieved by injecting noise in the system. We develop a theoretical model to understand this phenomenon and predict that it allows a hundred-fold energy gain over the synchronization of traditional dc-driven spin torque oscillators. Furthermore, we leverage our model to study the synchronization of several coupled junctions. Many theoretical schemes using the synchronization of oscillators to perform cognitive tasks such as pattern recognition and classification have been proposed. Using the noise-induced synchronization of superparamagnetic tunnel junctions would allow implementing these tasks at low energy.We draw an analogy between superparamagnetic tunnel junctions and sensory neurons which fire voltage pulses with random time intervals. Pushing this analogy, we demonstrate that populations of junctions can represent probability distributions and perform Bayesian inference. Furthermore, we demonstrate that interconnected populations can perform computing tasks such as learning, coordinate transformations and sensory fusion. Such a system is realistically implementable and could allow for intelligent sensory processing at low energy cost.

Dans cette thèse nous nous intéressons à la problématique de la construction de l'esthétiquechez les humains. Nous proposons d'utiliser un robot comme modèle pour étudier les briquesde bases qui participent au développement des préférences esthétiques. Nous utilisons le termed'esthétique artificielle (E.A ) pour désigner les préférences du robot.Plusieurs travaux de recherche tentent d'établir des théories de l'esthétique que nous séparons icien deux approches. D'une part, les approches empiriques qui étudientles préférences esthétiques d'un point de vue expérimental. Nous nous intéressons notamment àune branche plus radicale des approches empiriques, nommée la neuroesthétique. Celle-ci postulel'existence de structures cérébrales dédiées à l'appréciation des scènes visuelles en général et de l'art en particulier.D'autre part, les approches sociales qui avancent que les préférences esthétiques se transmettent de générationen génération et se construisent selon l'historique de l'individu et de ses interactions avec les autres.Le contextualisme historique est une branchedes approches sociales qui établit un lien entre le contexte dans lequel une œuvre est observée et son appréciation.Sans remettre en cause l'approche neuroscientifique, nous avons choisi de nous positionner dans une approche sociale et développementaleen utilisant des méthodes expérimentales telles que celles utilisées en esthétique empirique.Nous étudions l'émergence du sens esthétique dans le cadre de la référenciation sociale.On appelle référenciation sociale la capacité à attribuer des valences émotionnelles à des objets a priori neutre.Nous testons nos hypothèses sur robot mobile dans un cadre d'interaction triadique : homme-robot objet.Ceci dans un milieu naturel centré sur des humains non initiés à la robotique.Les humains jouent le rôle d'enseignants (professeur) du robot. Ils ont la tâche de suivre le robot dans son développementet de lui enseigner leurs préférences pour lui permettre de développer son propre "goût".Nous avons choisi de mener nos expériences dans des milieux dominés par l'esthétique comme les musées ou les galeries d'art.Toutefois, ces expériences peuvent être menées en tout lieu où des humains et des objets seraient disponibles.Notre robot, nommé Berenson en référence à un célèbre historien de l'art du 19ème siècle, est un outilpour comprendre d'une part comment s'installent des interactions sociales et comment les humainsprêtent des intentions aux machines, et d'autres part il permet d'étudier les briques minimalesd'intelligence artificielle à mettre en place pour construire une esthétique artificielle. / In this thesis we propose a robot as tool to study minimal bricks that helps human develop their aesthetic preferences. We refer to the robot preference using the term Artificial Esthetics (A.E).Several research work tries to establish a unified theory of esthetics. We divide them into two approaches. In one side, the empirical approaches which study esthetic preferences in an experimental manner. We mainly discuss the more radical branch of those approaches named "Neuroesthetic". Neuroesthetic advocates the existence of neural structures dedicated to visual scene preference and particularly to art appreciation. In another side, the social approaches which advocate that esthetic preferences are transmitted generation after generation, and they are built according to the individual historic and his interaction with others. Historical contextualism is a branch of the social approaches of art that draws a link between the appreciation of an artwork and the context where the artwork is observed.Without rejecting the neuroscientific approach, we choose a social and developmental way to study artificial esthetic using experimental methods from the empirical esthetic. We study the esthetic preferences development in the social referencing framework. Social referencing is the ability to attribute emotional values to à priori neutral objects. We test our hypothesis on a mobile robot in a triadic interaction : human-robot-object. This in a natural human centered environment. Humans play the role of the teachers. They have to fololow the robot in his development and teach it their preferences in order to help it develop its own "taste".We chose to conduct our experiment in places dominated by art and esthetics like museums and art galleries, however, this kind of experiment can take place anyway where human and objects are present.We named our robot Berenson in reference to a famous art historian of the 19th century. Berenson is a tool to understand how human project intentions into machines in one hand, and in the other hand the robot helps scientist build and understand minimal artificial intelligence bricks to build an artificial esthetic.