Sofisticadas técnicas para estimação de modelos baseadas em simulação, os filtros de partículas ou métodos de Monte Carlo Seqüenciais, foram empregadas recentemente para solucionar diversos problemas difícieis no campo da robótica móvel. No entanto, o sucesso dos fitros de partículas limitou-se à computação de parâmetros em espaços de baixa dimensionalidade. Os atuais esforços de pesquisa em robótica móvel têm comecado a explorar certas propriedades estruturais de seus domnios de aplicação que envolvem a utilização de filtros de partculas em espacos de maior dimensão, aumentando consideravelmente a complexidade da simulação envolvida. Simulações estatsticas dessa natureza requerem uma grande quantidade de numeros pseudo-aleatorios que possam ser gerados eficientemente e atendam a certos criterios de qualidade. O processo de geração de numeros pseudo-aleatorios torna-se o ponto crtico de tais aplicações em termos de desempenho. Neste contexto, a computação reconguravel insere-se como uma tecnologia capaz de satisfazer a demanda por alto desempenho das grandes simulações estatsticas pois sistemas baseados em arquiteturas reconguraveis possuem o potencial de mapear computação em hardware visando aumento de eficiência sem comprometer seriamente sua exibilidade. Tecnologias reconguraveis também possui o atrativo de um baixo consumo de energia, uma caracterstica essencial para os futuros robôs moveis embarcados. Esta dissertação apresenta a implementação um sistema embarcado baseado em FPGA e projetado para solucionar o problema de localização de robôs por meio de tecnicas probabilsticas. A parte fundamental de todo este sistema e um veloz gerador de numeros aleatorios mapeado ao hardware reconguravel que foi capaz de atender rígidos criterios estatsticos de qualidade / Sophisticated techniques for estimation of models based on simulation, particle filters or Sequential Monte Carlo Methods, were recently used to solve many difficult problems in the field of mobile robotics. However, the success of particle filters was limited to the computation of parameters in low dimensionality spaces. The current research efforts in mobile robotics have begun to explore some structural properties of their application\'s domain involving the use of particle filters in spaces of a higher dimension, greatly increasing the complexity of the involved simulation. Statistical simulations of this nature require a lot of pseudorandom numbers that can be generated efficiently and meet certain quality criteria. The process of generating pseudorandom number becomes the critical point of such applications in terms of performance. In this context, reconfigurable computing is a technology capable of meeting the demand for high performance of large statistical simulations because systems based on reconfigurable architectures have the potential to map computation to hardware aiming to increase eficiency without a serious drawback in exibility. Reconfigurable technologies are also attractive because of their low energy consume, a essential feature for the future mobile robots. This dissertation presents an implementation of a FPGA based embedded system designed to solve the robot localization problem by the means of probabilistic technics. The fundamental part from the whole system is a fast random number generator mapped to reconfigurable hardware wich atends a rigid quality criteria

Résumé non disponible / The race to computing power increases every day in the simulation community. A few years ago, scientists have started to harness the computing power of Graphics Processing Units (GPUs) to parallelize their simulations. As with any parallel architecture, not only the simulation model implementation has to be ported to the new parallel platform, but all the tools must be reimplemented as well. In the particular case of stochastic simulations, one of the major element of the implementation is the pseudorandom numbers source. Employing pseudorandom numbers in parallel applications is not a straightforward task, and it has to be done with caution in order not to introduce biases in the results of the simulation. This problematic has been studied since parallel architectures are available and is called pseudorandom stream distribution. While the literature is full of solutions to handle pseudorandom stream distribution on CPU-based parallel platforms, the young GPU programming community cannot display the same experience yet.In this thesis, we study how to correctly distribute pseudorandom streams on GPU. From the existing solutions, we identified a need for good software engineering solutions, coupled to sound theoretical choices in the implementation. We propose a set of guidelines to follow when a PRNG has to be ported to GPU, and put these advice into practice in a software library called ShoveRand. This library is used in a stochastic Polymer Folding model that we have implemented in C++/CUDA. Pseudorandom streams distribution on manycore architectures is also one of our concerns. It resulted in a contribution named TaskLocalRandom, which targets parallel Java applications using pseudorandom numbers and task frameworks.Eventually, we share a reflection on the methods to choose the right parallel platform for a given application. In this way, we propose to automatically build prototypes of the parallel application running on a wide set of architectures. This approach relies on existing software engineering tools from the Java and Scala community, most of them generating OpenCL source code from a high-level abstraction layer.

Sofisticadas técnicas para estimação de modelos baseadas em simulação, os filtros de partículas ou métodos de Monte Carlo Seqüenciais, foram empregadas recentemente para solucionar diversos problemas difícieis no campo da robótica móvel. No entanto, o sucesso dos fitros de partículas limitou-se à computação de parâmetros em espaços de baixa dimensionalidade. Os atuais esforços de pesquisa em robótica móvel têm comecado a explorar certas propriedades estruturais de seus domnios de aplicação que envolvem a utilização de filtros de partculas em espacos de maior dimensão, aumentando consideravelmente a complexidade da simulação envolvida. Simulações estatsticas dessa natureza requerem uma grande quantidade de numeros pseudo-aleatorios que possam ser gerados eficientemente e atendam a certos criterios de qualidade. O processo de geração de numeros pseudo-aleatorios torna-se o ponto crtico de tais aplicações em termos de desempenho. Neste contexto, a computação reconguravel insere-se como uma tecnologia capaz de satisfazer a demanda por alto desempenho das grandes simulações estatsticas pois sistemas baseados em arquiteturas reconguraveis possuem o potencial de mapear computação em hardware visando aumento de eficiência sem comprometer seriamente sua exibilidade. Tecnologias reconguraveis também possui o atrativo de um baixo consumo de energia, uma caracterstica essencial para os futuros robôs moveis embarcados. Esta dissertação apresenta a implementação um sistema embarcado baseado em FPGA e projetado para solucionar o problema de localização de robôs por meio de tecnicas probabilsticas. A parte fundamental de todo este sistema e um veloz gerador de numeros aleatorios mapeado ao hardware reconguravel que foi capaz de atender rígidos criterios estatsticos de qualidade / Sophisticated techniques for estimation of models based on simulation, particle filters or Sequential Monte Carlo Methods, were recently used to solve many difficult problems in the field of mobile robotics. However, the success of particle filters was limited to the computation of parameters in low dimensionality spaces. The current research efforts in mobile robotics have begun to explore some structural properties of their application\'s domain involving the use of particle filters in spaces of a higher dimension, greatly increasing the complexity of the involved simulation. Statistical simulations of this nature require a lot of pseudorandom numbers that can be generated efficiently and meet certain quality criteria. The process of generating pseudorandom number becomes the critical point of such applications in terms of performance. In this context, reconfigurable computing is a technology capable of meeting the demand for high performance of large statistical simulations because systems based on reconfigurable architectures have the potential to map computation to hardware aiming to increase eficiency without a serious drawback in exibility. Reconfigurable technologies are also attractive because of their low energy consume, a essential feature for the future mobile robots. This dissertation presents an implementation of a FPGA based embedded system designed to solve the robot localization problem by the means of probabilistic technics. The fundamental part from the whole system is a fast random number generator mapped to reconfigurable hardware wich atends a rigid quality criteria

The race to computing power increases every day in the simulation community. A few years ago, scientists have started to harness the computing power of Graphics Processing Units (GPUs) to parallelize their simulations. As with any parallel architecture, not only the simulation model implementation has to be ported to the new parallel platform, but all the tools must be reimplemented as well. In the particular case of stochastic simulations, one of the major element of the implementation is the pseudorandom numbers source. Employing pseudorandom numbers in parallel applications is not a straightforward task, and it has to be done with caution in order not to introduce biases in the results of the simulation. This problematic has been studied since parallel architectures are available and is called pseudorandom stream distribution. While the literature is full of solutions to handle pseudorandom stream distribution on CPU-based parallel platforms, the young GPU programming community cannot display the same experience yet. In this thesis, we study how to correctly distribute pseudorandom streams on GPU. From the existing solutions, we identified a need for good software engineering solutions, coupled to sound theoretical choices in the implementation. We propose a set of guidelines to follow when a PRNG has to be ported to GPU, and put these advice into practice in a software library called ShoveRand. This library is used in a stochastic Polymer Folding model that we have implemented in C++/CUDA. Pseudorandom streams distribution on manycore architectures is also one of our concerns. It resulted in a contribution named TaskLocalRandom, which targets parallel Java applications using pseudorandom numbers and task frameworks. Eventually, we share a reflection on the methods to choose the right parallel platform for a given application. In this way, we propose to automatically build prototypes of the parallel application running on a wide set of architectures. This approach relies on existing software engineering tools from the Java and Scala community, most of them generating OpenCL source code from a high-level abstraction layer.