Reactive motion planning and local navigation of robots remains a significant challenge in the motion control of robotic vehicles. This thesis presents new results on vision guided navigation using optical flow. By detecting key image features, calculating optical flow and leveraging time-to-transit (tau) as a feedback signal, control architectures can steer a vehicle so as to avoid obstacles while simultaneously using them as navigation beacons. Averaging and balancing tau over multiple image features successfully guides a vehicle along a corridor while avoiding looming objects in the periphery. In addition, the averaging strategy deemphasizes noise associated with rotationally induced flow fields, mitigating risks of positive feedback akin to the Larsen effect. A recently developed, biologically inspired, binary-key point description algorithm, FReaK, offers process speed-ups that make vision-based feedback signals achievable. A Parrot ARDrone2 has proven to be a reliable platform for testing the architecture and has demonstrated the control law's effectiveness in using time-to-transit calculations for real-time navigation.

Submitted by Patrícia Cerveira (pcerveira1@gmail.com) on 2017-06-14T20:26:32Z No. of bitstreams: 1 Alex João.pdf: 2737215 bytes, checksum: 01f1f6dd1595e3849310212c302e540e (MD5) / Approved for entry into archive by Biblioteca da Escola de Engenharia (bee@ndc.uff.br) on 2017-06-29T15:57:07Z (GMT) No. of bitstreams: 1 Alex João.pdf: 2737215 bytes, checksum: 01f1f6dd1595e3849310212c302e540e (MD5) / Made available in DSpace on 2017-06-29T15:57:07Z (GMT). No. of bitstreams: 1 Alex João.pdf: 2737215 bytes, checksum: 01f1f6dd1595e3849310212c302e540e (MD5) / O advento do paradigma financeiras, ou outros tipos de limitações. Desta forma, redes com enlaces redundantes têm sua capacidade potencial desperdiçada pelo uso de protocolos de roteamento distribuídos que não conseguem lidar de forma adequada com todos os caminhos disponíveis. Em especial, as redes dos centros de dados que possuem um grande número de enlaces redundantes e têm que fornecer a comunicação de forma eficiente entre seus equipamentos. Nesta dissertação será apresentado o Equalize, um arcabouço criado segundo o paradigma RDS, que realiza o equilíbrio automático dos fluxos ingressantes na rede através dos enlaces disponíveis que estejam menos congestionados. A topologia da rede é "descoberta" de forma automática e a localização de um host é determinada pelo seu endereço de camada 2 (MAC), de forma que o mesmo pode migrar para qualquer ponto da rede mantendo seu endereço de camada 3 (IP) original. Para o cálculo do melhor caminho entre dois nós, o Equalize executa um algoritmo de Dijkstra adaptado, que emprega a utilização de cada enlace como parâmetro de custo. Assim, o Equalize estabelece na rede a funcionalidade de auto-balanceamento e também apresenta uma boa resiliência para falhas, além de suportar a migração de hosts (sejam eles físicos ou virtuais). Com essas características, Equalize é bastante adequado para redes com facilidade para um gerenciamento centralizado e com múltiplos caminhos entre seus equipamentos, como as redes dos centros de dados. Os resultados obtidos com a emulação de uma pequena rede mostram que a abordagem proposta é capaz de alcançar uma boa taxa de transferência na bissecção da rede, ao mesmo tempo em que garante o equilíbrio no uso dos enlaces contribuindo, assim, para o uso eficiente dos recursos de rede e melhorando a utilização de sua capacidade total disponível. / The advent of Software-Defined Networking (SDN) is promoting innovation and faster network development. Even though network design demands for robust, more capable and flexible approaches to cope with increasingly complexity, system eficiency is often limited due to resource bounding as a consequence of physical, technical, financial, or other types of constraints. Though, networks with multiple redundant links have their potential capacity wasted due to the use of distributed routing protocols that cannot adequately deal with all available paths. In particular, the data center networks that have a large number of redundant links and must provide eficient communications among edges. In this dissertation we present Equalize, a framework built upon the SDN paradigm, that automatically balances the network incoming flows among the less congested available links. The network topology is automatically "discovered" and the location of a host is determined by its layer 2 (MAC) address, so that it can migrate to any point in the network while keeping its original layer 3 (IP) address. To calculate the best path between two nodes, Equalize runs an adapted Dijkstra's algorithm, employing each link utilization as the cost parameter. Thus, Equalize establishes the self-balancing functionality in the network and also presents a good resilience to failures, besides supporting the migration of hosts (whether physical or virtual). Having those characteristics, Equalize is well suited for networks with ease of centralized management and multiple paths among edges, such as data center networks. The results obtained by emulating a small network show that the proposed approach is able to achieve a good bisection throughput, while enforcing the balance in the use of links, thereby contributing to eficient use of network resources and improving the utilization of its total capacity available.