Este trabalho tem como objetivo o estudo de uma área da robótica chamada robótica móvel. Um robô móvel deve realizar uma navegação segura e esta é a principal motivação deste trabalho. Para tal foi desenvolvido um simulador de robótica móvel com visualização em 3D. Um dos grandes interesses na área de robótica móvel é a utilização de algoritmos de inteligência artificial. O objetivo deste trabalho é a utilização e simulação de inteligência artificial para o controle destinado ao desvio de obstáculos. As simulações são dinâmicas, ou seja, o robô não tem informação previa do cenário. Os algoritmos de inteligência artificial implementadas neste trabalho são lógica Fuzzy e Neurofuzzy. As contribuições do simulador são: a simulação e visualização em 3D com o cenário modelado em um programa CAD/3D, permite testar diversas configurações antes de testar o robô real, simula o ruído de sensores, utiliza lógica fuzzy e neurofuzzy para o desvio de obstáculos. Os resultados mostram a capacidade do sistema fuzzy para lidar com os dados ruidosos dos sensores assim como a influência das variáveis antecedentes e conseqüentes do sistema fuzzy de no comportamento do robô móvel para o desvio de obstáculos além da capacidade do sistema neurofuzzy de aprender a partir dos dados de treinamento mostrando uma melhoria no resultado das simulações.

In this thesis, contributions in the field of mobile robotics with a particular focus on assistive robotics are presented. The main subject addressed in this thesis is the navigation of robotic platforms taking into account the human component in robot decisions. For a robotic platform to navigate in an environment, effective sensor fusion and perception techniques are needed, in particular, accurate robot localization techniques and reliable representations of the obstacles surrounding a robot are required. Due to these requirements, this thesis was divided into two parts, corresponding to the two main areas of the research. The first part of the thesis focuses on the problem of localization and representation of the environment, and the second part focuses on the different levels of planning, namely: mission planning, global planning and local motion planning. As concerns the first part of the thesis, in order to improve robot localization in critical scenarios, where features of interest may not exist or be occluded, multimodal strategies are proposed, that use LiDAR, A-IPS and odometry, or only LiDAR and A-IPS, for localization using 2D occupancy grid maps. A map maintenance technique is also proposed that uses Reinforcement Learning (RL). Moreover, the HMap representation is also proposed, aiming to represent 3D environments with low computational and memory requirements, and fast access times. Methodologies for localization and Simultaneous Location and Mapping (SLAM) based on HMap, 2.5D and 3D representations are also presented. Contributions include different solutions for robot localization using 2D, 2.5D, 3D, HMap or Hybrid (2D and HMap) representations, as well as optimization methods to compute rigid transformations (3D and 6D) that minimize alignment errors between local and global maps, and loop closure detection methods. The results show that the proposed methods provide reliable robot localization and accurate representations in different test scenarios. As concerns the second part of the thesis, in the context of mission planning, the classic problem of constructing a Euclidean minimum Steiner tree, using soft obstacles is addressed. For the global planning, a method based on the Rapidly-Exploring Random Tree (RRT) is proposed, where dynamic obstacles are considered. In addition, as concerns local motion planning, cost maps are used in RL and Deep RL (DeepRL) approaches, with contributions in the architecture of the methods and in the training techniques. For the local motion planning, in the context of assistive robotics and following a user-centered navigation paradigm, where the user's intent represents an important element in the decision of actions, different methods for robotic walkers and wheelchairs are presented. Using RL-based methods, the main goal is to find the safest paths in order to guarantee the safety of the platform and the user. Learning from the user's behavior and thus preventing risky situations was also an important concern. These methods are particularly important when considering that the users of these assistive devices suffer from different types of mobility limitations and have difficulty in safely controlling or maneuvering the assistive platforms. The reported results show the capabilities and limitations of the RL and DeepRL methods in local motion planning applications. For the problem of local motion planning in the context of assistive robotics, tests with volunteers were carried out in order to validate the different proposed methods, achieving very promising results. The methodologies presented were validated using online datasets, datasets acquired at ISR-UC and tests performed by volunteers. Particular focus was placed on validation using robotic platforms in real scenarios, with the various modules developed from scratch in ROS. The use of platforms developed in the ISR-UC HcMR lab, such as the InterBot and the robotic walker ISR-AIWALKER, was essential for the field validation of the proposed methods.

This thesis addresses the problem of how to share efficiently information within a robotic system comprised of several mobile robots, which are programmed to exhibit cooperative behavior in the context of building volumetric maps of unknown environments. More specifically, it addresses the following issues: representing a probabilistic map and improving it through efficient exploration, based on information gain maximization; distributed control of teams of cooperative mobile robots, based on an information utility criteria; and coordinated exploration, aiming at avoiding redundant sensory information and robots' interference. Robots have been developed essentially to help or substitute humans in tasks which are either repetitive or dangerous. For many of these tasks, especially those that are intrinsically distributed and complex, a team of several cooperative mobile robots ─ a cooperative multi-robot system (MRS) ─ is required to either make viable the mission accomplishment or, at least, accomplish the mission with better performance than a single mobile robot. In spite of potential advantages related with space distribution, time distribution, complex problems decomposition, robustness, reliability and cost, a MRS requires that each robot maintains a sufficient and consistent level of awareness about the mission assigned to the team and about its teammates, in order to attain effective cooperation. The main challenge is that information is distributed and thus each robot has only partial and, sometimes, inconsistent knowledge about the environment. Sharing efficiently information via communication is thus crucial for robots' cooperation. Building maps is indeed a relevant robotics' application domain. Firstly, in many other application domains than building maps (e.g. search and rescue, surveillance, planetary exploration, etc.), a robot usually needs a map to support safe and efficient navigation based on a world model. Secondly, robots may substitute humans on building detailed models, such as: fastidious maps of indoor environments (e.g. buildings); maps of buried utilities (e.g.gas pipes, power or communication lines, etc.), in order to avoid getting too close to them in construction activities; or detailed maps of hazardous environments (e.g. abandoned underground mines, nuclear facilities, etc.), in order to support monitoring or maintenance procedures. The contributions of this thesis include a compact grid-based probabilistic representation model of a volumetric map, which allows to explicitly model uncertainty through the entropy concept. A frontier-based exploration method is also formulated using entropy, so that each robot uses its current map to select a new exploration viewpoint with maximum information gain. This probabilistic framework is used to devise a distributed architecture model for building volumetric maps with teams of cooperative mobile robots, whereby each robot is altruistically committed to share useful measurements with its teammates. The information sharing is based on the formal definition of a measure of information utility, developed upon the concepts of entropy and mutual information, whereby sensory data is as useful as it contributes to improve the robot's map. The distributed architecture is further refined with a mechanism to coordinate the exploration actions of different robots, thus improving the team's performance. The proposed methods were implemented in mobile robots, equipped with wireless communication and a stereo-vision system providing range measurements. These robots were used to carry out a set of experiments in a physical environment, which successfully validated the proposed framework and demonstrated the performance improvement yielded by the robots' cooperation. The results obtained with mobile robots were complemented with extensive computer simulations, which demonstrated those methods with varying team sizes.

There is an increasing number of mobile robot applications. The demanding of the Industry 4.0 pushes the robotic areas in the direction of the decision. The autonomous robots should actually decide the path according to the dynamic environment. In some cases, time requirements must also be attended and require fast path planning methods. This paper addresses a comparison between well-known path planning methods using a realistic simulator that handles the dynamic properties of robot models including sensors. The methodology is implemented in SimTwo that allows to compare the A* and RRT* algorithms in different scenarios with dynamic and real time constraint scenarios.

Mobile robots are applied to various industrial contexts, performing repetitive and highperformance tasks. One way of generating interest in the study of robotics in this context is through robotics competitions. The aim of this work is the development of a 3D mobile robotics simulator with hardware-in-the-loop capabilities. It includes developing models for standard components, such as time-of-flight sensors, wheel encoders, and direct current motors. The simulator interacts with development boards, programmed through Arduino-compatible libraries for communication with each robot component. By having the microcontrollers process each sensor’s output and determine the appropriate motor commands, the microcontroller’s limitations are present even during the simulation. The simulator contains different environments, where users have to complete challenges that require sensor data to be interpreted and motor commands to be calculated for different purposes, namely following walls, controlling the robot speed, and developing algorithms for completing the micromouse competition. A modification to the flood fill algorithm, commonly used in the micromouse competition, was proposed and analysed. It targets robots with a simple movement set, unable to perform turns while maintaining linear speed. The simulator was used in the RoboSTEM hackathon, where students were presented with the challenge environments and developed their solutions. It provided insights about the problems they were asked to solve and the simulator software itself.

Sistemas ground-truth são extremamente úteis no campo da robótica, fornecendo dados externos precisos que são usados como referência na validação de sistemas de localização embebidos em robôs móveis. No entanto, sistemas ground-truth geralmente apresentam um custo elevado. O presente trabalho visa o desenvolvimento de um sistema ground-truth de baixo custo capaz de determinar a localização e orientação de agentes móveis por meio de um sensor laser rangefinder e uma câmera monocular. A metodologia criada é aplicável a diversos tipos de robôs móveis terrestres, bastando que seja possível fixar ao robô um alvo cilíndrico com marcadores coloridos. O sistema e a metodologia desenvolvida foram implementados e validados através do SimTwo, um simulador realista. Os resultados obtidos pelo sistema proposto foram comparados com os dados reais (fornecidos pelo simulador), demonstrando que a solução desenvolvida atende aos objetivos propostos. Por fim, são ressaltadas possíveis melhorias a serem implementadas em trabalhos futuros.

Some industries have critical areas (dangerous or hazardous) where the presence of a human must be reduced or avoided. In some cases, there are areas where humans should be replaced by robots. The present work uses a robot with differential drive to scan an environment with known and unknown obstacles, defined in 3D simulation. It is important that the robot be able to make the right decisions about its way without the need of an operator. A solution to this challenge will be presented in this paper. The control application and its communication module with a simulator or a real robot are proposed. The robot can perform the scan, passing through all the waypoints arranged in a grid. The results are presented, showcasing the robot’s capacity to perform a viable trajectory without human intervention.

This paper addresses the development of a new localization system based on a security laser presented on most AGVs for safety reasons. An enhanced artificial beacons detection algorithm is applied with a combination of a Kalman filter and an outliers rejection method in order to increase the robustness and precision of the system. This new robust approach allows to implement such system in current AGVs. Real results in industrial environment validate the proposed methodology.

Throughout this paper, a competition created to enable an inter-connection between the academic and industrial paradigms is presented, using Open Hardware and Software. This competition is called Robot at Factory Lite and serves as a case study as an additional enrollment for students to apply knowledge in the fields of programming, perception, motion planning, task planning, autonomous robotic, among others.

In mobile robotics, since no requirements have been defined regarding accuracy for Battery Management Systems (BMS), standard approaches such as Open Circuit Voltage (OCV) and Coulomb Counting (CC) are usually applied, mostly due to the fact that employing more complicated estimation algorithms requires higher computing power; thus, the most advanced BMS algorithms reported in the literature are developed and verified by laboratory experiments using PC-based software. The objective of this paper is to describe the design of an autonomous and versatile embedded system based on an 8-bit microcontroller, where a Dual Coulomb Counting Extended Kalman Filter (DCC-EKF) algorithm for State of Charge (SOC) estimation is implemented; the developed prototype meets most of the constraints for BMSs reported in the literature, with an energy efficiency of 94% and an error of SOC accuracy that varies between 2% and 8% based on low-cost components