This thesis describes the fuzzy logic based control system for an automated guided vehicle ( AGV ) designed to navigate from one position and orientation to another while avoiding obstacles. A vehicle with an onboard computer system and a beacon based location system has been used to provide experimental confirmation of the methods proposed during this research. A simulation package has been written and used to test control techniques designed for the vehicle. A series of navigation rules based upon the vehicle's current position relative to its goal produce a fuzzy fit vector, the entries in which represent the relative importance of sets defined over all the possible output steering angles. This fuzzy fit vector is operated on by a new technique called rule spreading which ensures that all possible outputs have some activation. An obstacle avoidance controller operates from information about obstacles near to the vehicle. A method has been devised for generating obstacle avoidance sets depending on the size, shape and steering mechanism of a vehicle to enable their definition to accurately reflect the geometry and dynamic performance of the vehicle. Using a set of inhibitive rules the obstacle avoidance system compiles a mask vector which indicates the potential for a collision if each one of the possible output sets is chosen. The fuzzy fit vector is multiplied with the mask vector to produce a combined fit vector representing the relative importance of the output sets considering the demands of both navigation and obstacle avoidance. This is operated on by a newly developed windowing technique which prevents any conflicts produced by this combination leading to an undesirable output. The final fit vector is then defuzzified to give a demand steering angle for the vehicle. A separate fuzzy controller produces a demand velocity. In tests carried out in simulation and on the research vehicle it has been shown that the control system provides a successful guidance and obstacle avoidance scheme for an automated vehicle.

With modern day computer power, developing intelligent vehicles is fast becoming a reality. An Intelligent Vehicle is a vehicle equipped with sensors and computing that allow it to perceive the world around it, and to decide on appropriate action. Vision cameras are a good choice to sense the environment. One key task of the camera in an intelligent vehicle is to detect and localise the obstacles, which is the preparation of path planning. Stereo vision based obstacle detection is used in this research. It does not analyse semantic meaning of image features, but directly measures the 3-D coordinates of image pixels, and thus is suitable for obstacle detection in an unknown environment. In this research, a novel correlation based stereo vision method is developed which greatly improves its accuracy while maintaining its real-time performance. Since a vision system provides a large amount of data, extracting refined information may sometimes be complex. In obstacle detection tasks, the purpose is to distinguish the obstacle pixels from the ground pixels in the disparity image. V-Disparity image approach is used in this research to detect the ground plane, however this approach relies heavily on sufficient road features. In this research, a correlation method to locate the ground plane in the disparity image, even without significant road features, is developed. Moreover, traditional V-Disparity images have difficulties detecting non-flat ground, thus having limited applications. This research also develops a method to detect non-flat ground using V-Disparity images, thus greatly widening its application.

The estimation of the gas concentration (process-state) associated with a stationary or moving source using a sensing aerial vehicle (SAV) is considered. The dispersion from such a gaseous source into the ambient atmosphere is representative of an accidental or deliberate release of chemicals, or a release of gases from biological systems. Estimation of the concentration field provides a superior ability for source localization, assessment of possible adverse impacts, and eventual containment. The abstract and finite-dimensional approximation framework presented couples theoretical estimation and control with computational fluid dynamics methods. The gas dispersion (process) model is based on the advection-diffusion equation with variable eddy diffusivities and ambient winds. Cases are considered for a 2D and 3D domain. The state estimator is a modified Luenberger observer with a €�collocated€� filter gain that is parameterized by the position of the SAV. The process-state (concentration) estimator is based on a 2D and 3D adaptive, multigrid, multi-step finite-volume method. The grid is adapted with local refinement and coarsening during the process-state estimation in order to improve accuracy and efficiency. The motion dynamics of the SAV are incorporated into the spatial process and the SAV€™s guidance is directly linked to the performance of the state estimator. The computational model and the state estimator are coupled in the sense that grid-refinement is affected by the SAV repositioning, and the guidance laws of the SAV are affected by grid-refinement. Extensive numerical experiments serve to demonstrate the effectiveness of the coupled approach.

In order for autonomous vehicles to achieve life-long operation in outdoor environments, navigation systems must be able to cope with visual change---whether it's short term, such as variable lighting or weather conditions, or long term, such as different seasons. As a GPS is not always reliable, autonomous vehicles must be self sufficient with onboard sensors. This thesis examines the problem of localisation against a known map across extreme lighting and weather conditions using only a stereo camera as the primary sensor. The method presented departs from traditional techniques that blindly apply out-of-the-box interest-point detectors to all images of all places. This naive approach fails to take into account any prior knowledge that exists about the environment in which the robot is operating. Furthermore, the point-feature approach often fails when there are dramatic appearance changes, as associating low-level features such as corners or edges is extremely difficult and sometimes not possible. By leveraging knowledge of prior appearance, this thesis presents an unsupervised method for learning a set of distinctive and stable (i.e., stable under appearance changes) feature detectors that are unique to a specific place in the environment. In other words, we learn place-dependent feature detectors that enable vastly superior performance in terms of robustness in exchange for a reduced, but tolerable metric precision. By folding in a method for masking distracting objects in dynamic environments and examining a simple model for external illuminates, such as the sun, this thesis presents a robust localisation system that is able to achieve metric estimates from night-today or summer-to-winter conditions. Results are presented from various locations in the UK, including the Begbroke Science Park, Woodstock, Oxford, and central London.

Accurate and robust localisation is a fundamental aspect of any autonomous mobile robot. However, if these are to become widespread, it must also be available at low-cost. In this thesis, we develop a new approach to localisation using monocular cameras by leveraging a coloured 3D pointcloud prior of the environment, captured previously by a survey vehicle. We make no assumptions about the external conditions during the robot's traversal relative to those experienced by the survey vehicle, nor do we make any assumptions about their relative sensor configurations. Our method uses no extracted image features. Instead, it explicitly optimises for the pose which harmonises the information, in a Shannon sense, about the appearance of the scene from the captured images conditioned on the pose, with that of the prior. We use as our objective the Normalised Information Distance (NID), a true metric for information, and demonstrate as a consequence the robustness of our localisation formulation to illumination changes, occlusions and colourspace transformations. We present how, by construction of the joint distribution of the appearance of the scene from the prior and the live imagery, the gradients of the NID can be computed and how these can be used to efficiently solve our formulation using Quasi-Newton methods. In order to reliably identify any localisation failures, we present a new classifier using the local shape of the NID about the candidate pose and demonstrate the performance gains of the complete system from its use. Finally, we detail the development of a real-time capable implementation of our approach using commodity GPUs and demonstrate that it outperforms a high-grade, commercial GPS-aided INS on 57km of driving in central Oxford, over a range of different conditions, times of day and year.

This thesis is about the construction and automatic target-less calibration of a 3D laser sensor; this is then used to localise an autonomous vehicle without using other sensors. Two novel contributions to our knowledge of robotics are presented here. The first is an automatic calibration routine, which is capable of learning its calibration parameters using only data from a 3D laser scanner. Targets with known dimensions are not required, as has previously been the case. The second main contribution is a localisation algorithm, which uses the high quality data from the calibrated 3D laser scanner with trajectory information from an additional source to build maps of the environment. The vehicle subsequently localises itself within these maps, using the 3D laser sensor alone. Inaccurate laser data manifests itself as blurring when it is plotted in 3D space. The automatic calibration routine recognises that the environment has a true underlying structure to it, and expresses the amount of disorder in the measured laser points using a cost function based on the entropy of the 3D laser data. By optimising this quantity, we obtain the true calibration parameters for the system. We have quantified the accuracy of this algorithm by simulating a static environment from which we draw laser measurements with known calibration parameters. It was found that our calibration system converges to the true calibration values of the sensor. We also address the problem of robotic localisation, as a continuous problem, evaluating precisely the continuous trajectory that the robot has taken as well as the location of the robotic platform. Maps are constructed using the high accuracy data stream from the 3D laser, combining it with an odometry stream, to build high quality laser point cloud maps. The algorithm localises the robotic platform within these maps using a single 3D laser sensor. We vary our estimate of the vehicle's trajectory, treating the scans from the 3D laser and the location of the vehicle as continuous data streams, in a way that maximally aligns the 3D laser data and the map; this is achieved by optimising a cost function based on the Kernelised Rényi Distance. This procedure is typically computationally taxing; however, the computational complexity and computation time of the overall system have been reduced considerably using an efficient algorithm known as the Improved Fast Gauss Transform (IFGT), making the system viable even for large amounts of laser data. An additional speedup was achieved by calculating the Jacobian of the cost function, rearranging it to a form calculable using IFGT approximations. These efficiencies reduce the cost of evaluating the system to near real time. We evaluate the accuracy of our localisation system by comparing it to a DGPS stream as the best available source of ground truth. We show that our system performs more consistently than DGPS. This was especially prominent in regions where the line of sight to GPS satellites was obscured by trees. It was found that the accuracy of our system was comparable to that of the DGPS system.

Many modern entry guidance systems depend on predictions of atmospheric parameters, notably atmospheric density, in order to guide the entry vehicle to some desired final state. However, in highly dynamic atmospheric environments such as the Martian atmosphere, the density may vary by as much as 200% from predicted pre-entry trends. This high level of atmospheric density uncertainty can cause significant complications for entry guidance processes and may in extreme scenarios cause complete failure of the entry. In the face of this uncertainty, mission designers are compelled to apply large trajectory and design safety margins which typically drive the system design towards less efficient solutions with smaller delivered payloads. The margins necessary to combat the high levels of atmospheric uncertainty may even preclude scientifically interesting destinations or architecturally useful mission modes such as aerocapture. Aerocapture is a method for inserting a spacecraft into an orbit about a planetary body with an atmosphere without the need for significant propulsive maneuvers. This can reduce the required propellant and propulsion hardware for a given mission which lowers mission costs and increases the available payload fraction. However, large density dispersions have a particularly acute effect on aerocapture trajectories due to the interaction of the high required speeds and relatively low densities encountered at aerocapture altitudes. Therefore, while the potential system level benefits of aerocapture are great, so too are the risks associated with this mission mode in highly uncertain atmospheric environments such as Mars. Contemporary entry guidance systems utilize static atmospheric density models for trajectory prediction and control. These static models are unable to alter the fundamental nature of the underlying state equations which are used to predict atmospheric density. This limits both the fidelity and adaptive freedom of these models and forces the guidance system to retroactively correct for the density prediction errors after those errors have already impacted the trajectory. A new class of dynamic density estimator called a Plastic Ensemble Neural System (PENS) is introduced which is able to generate high fidelity, adaptable density forecast models by altering the underlying atmospheric state equations to better agree with observed atmospheric trends. A new construct called an ensemble echo is also introduced which creates an associative learning architecture, permitting PENS to evolve with increasing atmospheric exposure. The PENS estimator is applied to a numerical guidance system and the performance of the composite system is investigated with over 144,000 guided trajectory simulations. The results demonstrate that the PENS algorithm achieves significant reductions in both the required post-aerocapture performance, and the aerocapture failure rates relative to historical density estimators.

No description available.

In this thesis, we present three main contributions to laser-based detection and tracking of dynamic objects, from both a model-based point of view and a model-free point of view, with an emphasis on applications to autonomous driving. A segmentation-based detector is first proposed to provide an end-to-end detection of the classes car, pedestrian and bicyclist in 3D laser data amongst significant background clutter. We postulate that, for the particular classes considered, solving a binary classification task outperforms approaches that tackle the multi-class problem directly. This is confirmed using custom and third-party datasets gathered of urban street scenes. The sliding window approach to object detection, while ubiquitous in the Computer Vision community, is largely neglected in laser-based object detectors, possibly due to its perceived computational inefficiency. We give a second thought to this opinion in this thesis, and demonstrate that, by fully exploiting the sparsity of the problem, exhaustive window searching in 3D can be made efficient. We prove the mathematical equivalence between sparse convolution and voting, and devise an efficient algorithm to compute exactly the detection scores at all window locations, processing a complete Velodyne scan containing 100K points in less than half a second. Its superior performance is demonstrated on the KITTI dataset, and compares commensurably with state of the art vision approaches. A new model-free approach to detection and tracking of moving objects with a 2D lidar is then proposed aiming at detecting dynamic objects of arbitrary shapes and classes. Objects are modelled by a set of rigidly attached sample points along their boundaries whose positions are initialised with and updated by raw laser measurements, allowing a flexible, nonparametric representation. Dealing with raw laser points poses a significant challenge to data association. We propose a hierarchical approach, and present a new variant of the well-known Joint Compatibility Branch and Bound algorithm to handle large numbers of measurements. The system is systematically calibrated on real world data containing 7.5K labelled object examples and validated on 6K test cases. Its performance is demonstrated over an existing industry standard targeted at the same problem domain as well as a classical approach to model-free tracking.

The past decade has seen great progress in research on large scale mapping and perception in static environments. Real world perception requires handling uncertain situations with multiple possible interpretations: e.g. changing appearances, dynamic objects, and varying motion models. These aspects of perception have been largely avoided through the use of heuristics and preprocessing. This thesis is motivated by the challenge of including discrete reasoning directly into the estimation process. We approach the problem by using Conditional Linear Gaussian Networks (CLGNs) as a generalization of least-squares estimation which allows the inclusion of discrete model selection variables. CLGNs are a powerful framework for modeling sparse multi-modal inference problems, but are difficult to solve efficiently. We propose the Iterative Local Model Selection (ILMS) algorithm as a general approximation strategy specifically geared towards the large scale problems encountered in tracking and mapping. Chapter 4 introduces the ILMS algorithm and compares its performance to traditional approximate inference techniques for Switching Linear Dynamical Systems (SLDSs). These evaluations validate the characteristics of the algorithm which make it particularly attractive for applications in robot perception. Chief among these is reliability of convergence, consistent performance, and a reasonable trade off between accuracy and efficiency. In Chapter 5, we show how the data association problem in multi-target tracking can be formulated as an SLDS and effectively solved using ILMS. The SLDS formulation allows the addition of additional discrete variables which model outliers and clutter in the scene. Evaluations on standard pedestrian tracking sequences demonstrates performance competitive with the state of the art. Chapter 6 applies the ILMS algorithm to robust pose graph estimation. A non-linear CLGN is constructed by introducing outlier indicator variables for all loop closures. The standard Gauss-Newton optimization algorithm is modified to use ILMS as an inference algorithm in between linearizations. Experiments demonstrate a large improvement over state-of-the-art robust techniques. The ILMS strategy presented in this thesis is simple and general, but still works surprisingly well. We argue that these properties are encouraging for wider applicability to problems in robot perception.