Many mobile robot applications require robots to act safely and intelligently in complex unfamiliarenvironments with little structure and limited or unavailable human supervision. As arobot is forced to operate in an environment that it was not engineered or trained for, various aspectsof its performance will inevitably degrade. Roboticists equip robots with powerful sensorsand data sources to deal with uncertainty, only to discover that the robots are able to make onlyminimal use of this data and still find themselves in trouble. Similarly, roboticists develop andtrain their robots in representative areas, only to discover that they encounter new situations thatare not in their experience base. Small problems resulting in mildly sub-optimal performance areoften tolerable, but major failures resulting in vehicle loss or compromised human safety are not.This thesis presents a series of online algorithms to enable a mobile robot to better deal withuncertainty in unfamiliar domains in order to improve its navigational abilities, better utilizeavailable data and resources and reduce risk to the vehicle. We validate these algorithms throughextensive testing onboard large mobile robot systems and argue how such approaches can increasethe reliability and robustness of mobile robots, bringing them closer to the capabilitiesrequired for many real-world applications.

Advances in hardware design have made wheeled mobile robots (WMRs) exceptionally mobile. To fully exploit this mobility, WMR planning, control, and estimation systems require motion models that are fast and accurate. Much of the published theory on WMR modeling is limited to 2D or kinematics, but 3D dynamic (or force-driven) models are required when traversing challenging terrain, executing aggressive maneuvers, and manipulating heavy payloads. This thesis advances the state of the art in both the formulation and calibration of WMR models We present novel WMR model formulations that are high-fidelity, general, modular, and fast. We provide a general method to derive 3D velocity kinematics for any WMR joint configuration. Using this method, we obtain constraints on wheel ground contact point velocities for our differential algebraic equation (DAE)-based models. Our “stabilized DAE” kinematics formulation enables constrained, drift free motion prediction on rough terrain. We also enhance the kinematics to predict nonzero wheel slip in a principled way based on gravitational, inertial, and dissipative forces. Unlike ordinary differential equation (ODE)-based dynamic models which can be very stiff, our constrained dynamics formulation permits large integration steps without compromising stability. Some alternatives like Open Dynamics Engine also use constraints, but can only approximate Coulomb friction at contacts. In contrast, we can enforce realistic, nonlinear models of wheel-terrain interaction (e.g. empirical models for pneumatic tires, terramechanics-based models) using a novel force-balance optimization technique. Simulation tests show our kinematic and dynamic models to be more functional, stable, and efficient than common alternatives. Simulations run 1K-10K faster than real time on an ordinary PC, even while predicting articulated motion on rough terrain and enforcing realistic wheel-terrain interaction models. In addition, we present a novel Integrated Prediction Error Minimization (IPEM) method to calibrate model parameters that is general, convenient, online, and evaluative. Ordinarily system dynamics are calibrated by minimizing the error of instantaneous output predictions. IPEM instead forms predictions by integrating the system dynamics over an interval; benefits include reduced sensing requirements, better observability, and accuracy over a longer horizon. In addition to calibrating out systematic errors, we simultaneously calibrate a model of stochastic error propagation to quantify the uncertainty of motion predictions. Experimental results on multiple platforms and terrain types show that parameter estimates converge quickly during online calibration, and uncertainty is well characterized. Under normal conditions, our enhanced kinematic model can predict nonzero wheel slip as accurately as a full dynamic model for a fraction of the computation cost. Finally, odometry is greatly improved when using IPEM vs. manual calibration, and when using 3D vs. 2D kinematics. To facilitate their use, we have released open source MATLAB and C++ libraries implementing the model formulation and calibration methods in this thesis.

A new hybrid micro vehicle, called the Hopping Rotochute, was developed to robustly explore environments with rough terrain while minimizing energy consumption over extended periods of time. Unlike traditional robots, the Hopping Rotochute maneuvers through complex terrain by hopping over or through impeding obstacles. A small coaxial rotor system provides the necessary lift while a movable internal mass controls the direction of travel. In addition, the low mass center and egg-like shaped body creates a means to passively reorient the vehicle to an upright attitude when in ground contact while protecting the rotating components. The design, fabrication, and testing of a radio-controlled Hopping Rotochute prototype as well as an analytical study of the flight performance are documented. The aerodynamic, mechanical, and electrical design of the prototype is outlined which were driven by the operational requirements assigned to the vehicle. The aerodynamic characteristics of the rotor system as well as the damping characteristics of the foam base are given based on experimental results using a rotor test stand and a drop test stand respectively. Experimental flight testing results using the prototype are outlined which demonstrate that all design and operational requirements are satisfied. A dynamic model associated with the Hopping Rotochute is then developed including a soft contact model which estimates the forces and moments on the vehicle during ground contact. A comparison between the vehicle's motion measured using a motion capture system and the simulation results are presented to determine the validity of the experimentally-tuned dynamic model. Using this validated simulation model, key parameters such as system weight, rotor speed profile, internal mass weight and location, as well as battery capacity are varied to explore the flight performance characteristics. The sensitivity of the hopping rotochute to atmospheric winds is also investigated as well as the ability of the device to perform trajectory shaping.

This thesis investigates the use of visual-motion information sampled through optical flow for the task of indoor obstacle avoidance on autonomous mobile robots. The methods focus on the practical use of optical flow and visual motion information in performing the obstacle avoidance task in real indoor environments. The methods serve to identify visual-motion properties that must be used in synergy with visual-spatial properties toward the goal of a complete robust visual-only obstacle avoidance system, as is evidently seen within nature. A review of vision-based obstacle avoidance techniques shows that early research mainly focused on visual-spatial techniques, which heavily rely on various assumptions of their environments to function successfully. On the other hand, more current research that looks toward the use of visual-motion information (sampled through optical flow) tends to focus on using optical flow in a subsidiary manner, and does not completely take advantage of the information encoded within an optical flow field. In the light of the current research limitations, this thesis describes two different approaches and evaluates their use of optical flow to perform the obstacle avoidance task. The first approach begins with the construction of a conventional range map using optical flow that stems from the structure-from-motion domain and the theory that optical flow encodes 3D environmental information under certain conditions. The second approach investigates optical flow in a causal mechanistic manner using machine learning of motor responses directly from optical flow - motivated from physical and behavioural evidence observed in biological creatures. Specifically, the second approach is designed with three main objectives in mind: 1) to investigate whether optical flow can be learnt for obstacle avoidance; 2) to create a system capable of repeatable obstacle avoidance performance in real-life environments; and 3) to analyse the system to determine what optical flow properties are actually being used for the motor control task. The range-map reconstruction results have demonstrated some good distance estimations through the use of a feature-based optical flow algorithm. However, the number of flow points were too sparse to provide adequate obstacle detection. Results froma differential-based optical flow algorithm helped to increase the density of flow points, but highlighted the high sensitivity of the optical flow field to the rotational errors and outliers that plague the majority of frames under real-life robot situations. Final results demonstrated that current optical flow algorithms are ill-suited to estimate obstacle distances consistently, as range-estimation techniques require an extremely accurate optical flow field with adequate density and coverage for success. This is a difficult problem within the optical flow estimation domain itself. In the machine learning approach, an initial study to examine whether optical flow can be machine learnt for obstacle avoidance and control in a simple environment was successful. However,there were certain problems. Several critical issues which arise with the use of a machine learning approach were highlighted. These included sample set completeness, sample set biases, and control system instability. Consequently, an extended neural network was proposed that had several improvements made to overcome the initial problems. Designing an automated system for gathering training data helped to eliminate most of the sample set problems. Key changes in the neural network architecture, optical flow filters, and navigation technique vastly improved the control system stability. As a result, the extended neural network system was able to successfully perform multiple obstacle avoidance loops in both familiar and unfamiliar real-life environments without collisions. The lap times of the machine learning approach were comparable to those of the laser-based navigation technique. The the machine learning approach was 13% slower in the familiar and 25% slower in the unfamiliar environment. Furthermore, through analysis of the neural network approach, flow magnitudes were revealed to be learnt for range information in an absolute manner, while flow directions were used to detect the focus of expansion (FOE) in order to predict critical collision situations and improve control stability. In addition, the precision of the flow fields was highlighted as an important requirement, as opposed to the high accuracy of flow vectors. For robot control purposes, image-processing techniques such as region finding and object boundary detection were employed to detect changes between optical flow vectors in the image space.

This thesis investigates the use of visual-motion information sampled through optical flow for the task of indoor obstacle avoidance on autonomous mobile robots. The methods focus on the practical use of optical flow and visual motion information in performing the obstacle avoidance task in real indoor environments. The methods serve to identify visual-motion properties that must be used in synergy with visual-spatial properties toward the goal of a complete robust visual-only obstacle avoidance system, as is evidently seen within nature. A review of vision-based obstacle avoidance techniques shows that early research mainly focused on visual-spatial techniques, which heavily rely on various assumptions of their environments to function successfully. On the other hand, more current research that looks toward the use of visual-motion information (sampled through optical flow) tends to focus on using optical flow in a subsidiary manner, and does not completely take advantage of the information encoded within an optical flow field. In the light of the current research limitations, this thesis describes two different approaches and evaluates their use of optical flow to perform the obstacle avoidance task. The first approach begins with the construction of a conventional range map using optical flow that stems from the structure-from-motion domain and the theory that optical flow encodes 3D environmental information under certain conditions. The second approach investigates optical flow in a causal mechanistic manner using machine learning of motor responses directly from optical flow - motivated from physical and behavioural evidence observed in biological creatures. Specifically, the second approach is designed with three main objectives in mind: 1) to investigate whether optical flow can be learnt for obstacle avoidance; 2) to create a system capable of repeatable obstacle avoidance performance in real-life environments; and 3) to analyse the system to determine what optical flow properties are actually being used for the motor control task. The range-map reconstruction results have demonstrated some good distance estimations through the use of a feature-based optical flow algorithm. However, the number of flow points were too sparse to provide adequate obstacle detection. Results froma differential-based optical flow algorithm helped to increase the density of flow points, but highlighted the high sensitivity of the optical flow field to the rotational errors and outliers that plague the majority of frames under real-life robot situations. Final results demonstrated that current optical flow algorithms are ill-suited to estimate obstacle distances consistently, as range-estimation techniques require an extremely accurate optical flow field with adequate density and coverage for success. This is a difficult problem within the optical flow estimation domain itself. In the machine learning approach, an initial study to examine whether optical flow can be machine learnt for obstacle avoidance and control in a simple environment was successful. However,there were certain problems. Several critical issues which arise with the use of a machine learning approach were highlighted. These included sample set completeness, sample set biases, and control system instability. Consequently, an extended neural network was proposed that had several improvements made to overcome the initial problems. Designing an automated system for gathering training data helped to eliminate most of the sample set problems. Key changes in the neural network architecture, optical flow filters, and navigation technique vastly improved the control system stability. As a result, the extended neural network system was able to successfully perform multiple obstacle avoidance loops in both familiar and unfamiliar real-life environments without collisions. The lap times of the machine learning approach were comparable to those of the laser-based navigation technique. The the machine learning approach was 13% slower in the familiar and 25% slower in the unfamiliar environment. Furthermore, through analysis of the neural network approach, flow magnitudes were revealed to be learnt for range information in an absolute manner, while flow directions were used to detect the focus of expansion (FOE) in order to predict critical collision situations and improve control stability. In addition, the precision of the flow fields was highlighted as an important requirement, as opposed to the high accuracy of flow vectors. For robot control purposes, image-processing techniques such as region finding and object boundary detection were employed to detect changes between optical flow vectors in the image space.

Thesis (M.Sc. [i.e. M.C.M.S.])--University of Waikato, 2006. / Includes bibliographical references (p. 167-173) Also available via the World Wide Web.

Com a necessidade de substituir o trabalho humano em áreas de risco, ambientes impróprios ou inalcançáveis, diversos centros de pesquisas e universidades têm desenvolvido aplicações e estratégias de controle para robôs móveis. Porém, o alto custo na aquisição de um ou mais protótipos para estudos e desenvolvimento de novas tecnologias pode se tornar um fator limitante para o incremento dessas pesquisas. Como parte de uma solução para contornar esta eventual limitação em robótica móvel, uma arquitetura de baixo custo, modular e expansível é apresentada neste trabalho. São apresentadas também as metodologias de desenvolvimento dos módulos, os algoritmos de controle, as interfaces de comunicação e os principais componentes utilizados para desenvolvimento do robô móvel ZEUS, cujo sistema eletrônico de controle é a implementação da arquitetura proposta. Análise de custo, resultados experimentais de sensoriamento e navegação concluem este trabalho. / With the need to replace human work in risk\'s areas, improper or unreachable environments, several research centers, and universities have developed applications and strategies for mobile robots control. However, the high acquisition cost of one or more prototypes used for research and in development of new technologies may become a limiting factor. As part of the solution to get a round any such limitation on mobile robotics, a modular and expandable low-cost architecture is presented in this work. There are also presented the methodologies for modules development, the algorithms of control, the communication interfaces and the main components used for the development of ZEUS mobile robot, whose electronic system control is the implementation of the proposed architecture. Analysis of cost, experimental results of the sensing and the navigation system conclude this work.

Planejamento de movimento tem o propósito de determinar quais movimentos o robô deve realizar para que alcance posições ou configurações desejadas no ambiente sem que ocorram colisões com obstáculos. É comum na robótica móvel simplificar o planejamento de movimento representando o robô pelas coordenadas do seu centro e desconsiderando qualquer restrição cinemática e dinâmica de movimento. Entretanto, a maioria dos robôs móveis possuem restrições cinemáticas não-holonômicas, e para algumas tarefas e robôs, é importante considerar tais restrições juntamente com o modelo dinâmico do robô na tarefa de planejamento. Assim é possível determinar um caminho que possa ser de fato seguido pelo robô. Nesse trabalho é proposto um método de planejamento cinemático-dinâmico que permite planejar trajetórias para robôs móveis usando malhas de estados. Essa abordagem considera a cinemática e a dinâmica do robô para gerar trajetórias possíveis de serem executadas e livre de colisões com obstáculos. Quando obstáculos não representados no mapa são detectados pelos sensores do robô, uma nova trajetória é gerada para desviar desses obstáculos. O planejamento de movimento utilizando malhas de estados foi associado a um algoritmo de desvio de obstáculos baseado no método da janela dinâmica (DWA). Esse método é responsável pelo controle de seguimento de trajetória, garantindo a segurança na realização da tarefa durante a navegação. As trajetórias planejadas foram executadas em duas plataformas distintas. Essas plataformas foram utilizadas em tarefas de navegação em ambientes simulados interno e externo e em ambientes reais. Para navegação em ambientes internos utilizou-se o robô móvel Pioneer 3AT e para navegação em ambientes externos utilizou-se o veículo autônomo elétrico CaRINA 1 que está sendo desenvolvido no ICMC-USP com apoio do Instituto Nacional de Ciência e Tecnologia em Sistemas Embarcados Críticos (INCT-SEC). / Motion planning aims to determine which movements the robot must accomplish to reach a desired position or configuration in the environment without the occurrence of collisions with obstacles. It is common in mobile robotics to simplify the motion planning representing the robot by the coordinates of its center of gravity and ignoring any kinematic and dynamic constraint motion. However, most mobile robots have non-holonomic kinematic constraints, and for some tasks and robots, it is important to consider these constraints together with the dynamic model of the robot in task planning. Thus it is possible to determine a path that can actually be followed by the robot. Here we propose a method for kinematic-dynamic path planning using lattice states. This approach considers the kinematic and dynamic of the robot to generate generate feasible trajectories free of collisions with obstacles. When obstacles not represented on the map are detected by the sensors of the robot, a new trajectory is generated to avoid these obstacles. The motion planning using lattice state was associated with an obstacle avoidance algorithm based on the dynamic window approach (DWA). This method is responsible for trajectory tracking to ensure safety in navigation tasks. This method was applied in two distinct platforms. These platforms were used for navigation tasks in both indoor and outdoor simulated environments, as well as, in real environments. For navigation in indoor environments we used a Pioneer 3AT robot and for outdoor navigation we used the autonomous electric vehicle CaRINA1 being developed at ICMC-USP with support National Institute of Science and Technology in Critical Embedded Systems (INCT-SEC).

Submitted by DIEGO RENAN BRUNO null (diego_renan_bruno@hotmail.com) on 2016-05-19T02:22:22Z No. of bitstreams: 1 DISSERTAÇÃO_Diego_Renan_Bruno_UNESP_IBILCE.pdf: 5786200 bytes, checksum: e4c11c8581ba4cc8af49f068d9d637d9 (MD5) / Approved for entry into archive by Felipe Augusto Arakaki (arakaki@reitoria.unesp.br) on 2016-05-23T14:27:37Z (GMT) No. of bitstreams: 1 bruno_dr_me_sjrp.pdf: 5786200 bytes, checksum: e4c11c8581ba4cc8af49f068d9d637d9 (MD5) / Made available in DSpace on 2016-05-23T14:27:37Z (GMT). No. of bitstreams: 1 bruno_dr_me_sjrp.pdf: 5786200 bytes, checksum: e4c11c8581ba4cc8af49f068d9d637d9 (MD5) Previous issue date: 2016-04-27 / Neste trabalho foi desenvolvido um sistema de controle híbrido bioinspirado para o planejamento de rota com vistas para a robótica móvel autônoma, baseado em redes neurais artificiais e algoritmos genéticos. O controlador tem como principal objetivo auxiliar o robô móvel em sua navegação quando aplicado em ambientes dinâmicos. Para este trabalho, o ambiente dinâmico utilizado é um “chão de fábrica” industrial, em que alguns obstáculos não são fixos e permanecem em movimentação constante. O controlador desenvolvido neste trabalho pode ser adaptado facilmente para operar em outros ambientes dinâmicos. Independentemente do ambiente utilizado, o controlador deve ser capaz de traçar uma rota possível entre o ponto inicial e o ponto de objetivo, tendo o potencial de evitar todo tipo de obstáculo que surgir nessa rota, seja um obstáculo estático ou dinâmico. O algoritmo foi implementado na linguagem C e simulado no software de modelagem e simulação de robôs V-REP (Virtual Robot Experimentation Platform). O controlador neurogenético mostrou ser eficiente para auxiliar o robô em sua navegação quando aplicado em ambientes dinâmicos. / In this work, a bioinspired hybrid control system was developed for route planning, aiming autonomous mobile robots based on artificial neural networks and genetic algorithms. The main objective of the controller is to assist the mobile robot in its navigation when applied in dynamic environments. For this work, the dynamic environment chosen was a " factory floor", in which some industrial obstacles are not fixed and remain in constant movements. The controller developed in this work can easily be adapted to operate in other dynamic environments. Regardless the environment chosen in this work, the controller must be able to map out a possible route between the starting point and the goal point with the potential to avoid all types of obstacles that appear along the routes, either a static or a dynamic one. The algorithm was implemented in C language and simulated on a robots modeling and simulation software called V-REP (Virtual Robot Experimentation Platform). The neurogenetic controller was efficient to assist the mobile robot in its navigation when applied in dynamic environments.

In order to improve the comfort of power wheelchair users, it is proposed in this dissertation the implementation of a position and orientation control system which allow the users to travel the desired trajectories safely, without the need for constant interaction with the device command system. To reach this objective, it is proposed a methodology for the implementation of a control system, able to drive the vehicle from an initial position to the desired one or make it follow a specified trajectory, from commands received in the system. The proposed control system is based on a cascade control law, composed by a nonlinear pose controller and a velocity tracking controller. Two velocity control strategies are proposed. The first one uses a velocity controller composed of two joint-independent Proportional-Integral controllers while the second one uses a velocity controller based on the dynamic model, which is linearized by a state-feedback. The methods are implemented using the ros_control package, provided by the framework Robot Operating System (ROS). The evaluation of the proposed methods is done with a differential-drive mobile robot, which has the same kinematic configuration as the majority of commercial power wheelchairs. The results of dynamic-model parameter identification, as well as the convergence of the controlled variables by using the proposed control methods, are presented. The results demonstrate that the methods achieve the proposed control objectives.