This thesis aims to design an appropriate human-following solution for a mobile robot. The research can be characterised as interactive model building for a Human Robot Interaction (HRI) scenario. It studies possible proposals for the robot system that learns to accomplish the task autonomously, based on the human preference about the positions and movements of the robot during the interaction. A multilayered feedforward network framework with backpropagation is the adopted learning strategy. The research breaks the task of following a human into three independent behaviours: social positioning, human avoidance and obstacle avoidance. Social positioning is the behaviour that moves the robot, via reasonable paths, to the most appropriate location to follow the human. Both the location and the paths reflect the preference of the human, which varies by individual. The main body of the research therefore proposes a using-while-learning system for this behaviour such that the robot can adapt to the human’s preference autonomously. This research investigated multilayered feedforward networks with backpropagation learning to fulfil the social learning task. This learning model is less used in HRI because a complete set of correct training data doesn’t exist as the human preference is initially unknown. The research proposes a novel method to generate the training data during the operation of learning and introduces the concept of adaptive and reactive learning. A novel training scheme that combines the two learning threads has been proposed, in which the learning is fast, robust and able to adapt to new features of the human preference online. The system enables the behaviour to be a real using-while-learning system as no pre-training of any form is needed to ensure the successful performance of the behaviour. Extensive simulations and interactive experiments with humans have also been conducted to prove the robustness of the system.

Urban Search and Rescue is a dangerous task for rescue workers and for this reason the use of mobile robots to carry out the search of the environment is becoming common place. These robots are remotely operated and the search is carried out by the robot operator. This work proposes that common search algorithms can be used to guide a single autonomous mobile robot in a search of an environment and locate survivors within the environment. This work then goes on to propose that multiple robots, guided by the same search algorithms, will carry out this task in a quicker time. The work presented is split into three distinct parts. The first is the development of a nonlinear mathematical model for a mobile robot. The model developed is validated against a physical system. A suitable navigation and control system is required to direct the robot to a target point within an environment. This is the second part of this work. The final part of this work presents the search algorithms used. The search algorithms generate the target points which allow the robot to search the environment. These algorithms are based on traditional and modern search algorithms that will enable a single mobile robot to search an area autonomously. The best performing algorithms from the single robot case are then adapted to a multi robot case. The mathematical model presented in the thesis describes the dynamics and kinematics of a four wheeled mobile ground based robot. The model is developed to allow the design and testing of control algorithms offline. With the model and accompanying simulation the search algorithms can be quickly and repeatedly tested without practical installation. The mathematical model is used as the basis of design for the manoeuvring control algorithm and the search algorithms. This design process is based on simulation studies. In the first instance the control methods investigated are Proportional-Integral-Derivative, Pole Placement and Sliding Mode. Each method is compared using the tracking error, the steady state error, the rise time, the charge drawn from the battery and the ability to control the robot through a simple motion. Obstacle avoidance is also covered as part of the manoeuvring control algorithm. The final aspect investigated is the search algorithms. The following search algorithms are investigated, Lawnmower, Random, HillClimbing, Simulated Annealing and Genetic Algorithms. Variations on these algorithms are also investigated. The variations are based on Tabu Search. Each of the algorithms is investigated in a single robot case with the best performing investigated within a multi robot case. A comparison between the different methods is made based on the percentage of the area covered within the time available, the number of targets located and the time taken to locate targets. It is shown that in the single robot case the best performing algorithms have high random elements and some structure to selecting points. Within the multi robot case it is shown that some algorithms work well and others do not. It is also shown that the useable number of robots is dependent on the size of the environment. This thesis concludes with a discussion on the best control and search algorithms, as indicated by the results, for guiding single and multiple autonomous mobile robots. The advantages of the methods are presented, as are the issues with using the methods stated. Suggestions for further work are also presented.

For a mobile robot to engage in exploration of a-priori unknown environments it must be able to identify locations which will yield new information when visited. This thesis presents two novel algorithms which attempt to answer the question of choosing where a robot should go next in a partially explored workspace. To begin we describe the process of acquiring highly accurate dense 3D data from a stereo camera. This approach combines techniques from a number of existing implementations and is demonstrated to be more accurate than a range of commercial offerings. Combined with state of the art visual odometry based pose estimation we can use these point clouds to drive exploration. The first exploration algorithm we present is an attempt to represent the three dimensional world as a continuous two dimensional surface. The surface is maintained as a planar graph structure in which vertices correspond to points in space as seen by the stereo camera. Edges connect vertices which have been seen as adjacent pixels in a stereo image pair, and have a weight equal to the Euclidean distance between the end points. Discontinuities in the input stereo data manifest as areas of the graph with high average edge weight, and by moving the camera to view such areas and merging the new scan with the existing graph, we push back the boundary of the explored workspace. Motivated by scaling and precision problems with the graph-based method, we present a second exploration algorithm based on continuum methods. We show that by solving Laplace’s equation over the freespace of the partially explored environment, we can guide exploration by following streamlines in the resulting vector field. Choosing appropriate boundary conditions ensures that these streamlines run parallel to obstacles and are guaranteed to lead to a frontier – a boundary between explored and unexplored space. Results are shown which demonstrate this method fully exploring three dimensional environments and outperforming oft-used information gain based approaches. We show how analysis of the potential field solution can be used to identify volumes of the workspace which have been fully explored, thus reducing future computation.

In this thesis, a cognitive neural predictive controller system has been designed to guide a nonholonomic wheeled mobile robot during continuous and non-continuous trajectory tracking and to navigate through static obstacles with collision-free and minimum tracking error. The structure of the controller consists of two layers; the first layer is a neural network system that controls the mobile robot actuators in order to track a desired path. The second layer of the controller is cognitive layer that collects information from the environment and plans the optimal path. In addition to this, it detects if there is any obstacle in the path so it can be avoided by re-planning the trajectory using particle swarm optimisation (PSO) technique. Two neural networks models are used: the first model is modified Elman recurrent neural network model that describes the kinematic and dynamic model of the mobile robot and it is trained off-line and on-line stages to guarantee that the outputs of the model will accurately represent the actual outputs of the mobile robot system. The trained neural model acts as the position and orientation identifier. The second model is feedforward multi-layer perceptron neural network that describes a feedforward neural controller and it is trained off-line and its weights are adapted on-line to find the reference torques, which controls the steady-state outputs of the mobile robot system. The feedback neural controller is based on the posture neural identifier and quadratic performance index predictive optimisation algorithm for N step-ahead prediction in order to find the optimal torque action in the transient to stabilise the tracking error of the mobile robot system when the trajectory of the robot is drifted from the desired path during transient state. Three controller methodologies were developed: the first is the feedback neural controller; the second is the nonlinear PID neural feedback controller and the third is nonlinear inverse dynamic neural feedback controller, based on the back-stepping method and Lyapunov criterion. The main advantages of the presented approaches are to plan an optimal path for itself avoiding obstructions by using intelligent (PSO) technique as well as the analytically derived control law, which has significantly high computational accuracy with predictive optimisation technique to obtain the optimal torques control action and lead to minimum tracking error of the mobile robot for different types of trajectories. The proposed control algorithm has been applied to monitor a nonholonomic wheeled mobile robot, has demonstrated the capability of tracking different trajectories with continuous gradients (lemniscates and circular) or non-continuous gradients (square) with bounded external disturbances and static obstacles. Simulations results and experimental work showed the effectiveness of the proposed cognitive neural predictive control algorithm; this is demonstrated by the minimised tracking error to less than (1 cm) and obtained smoothness of the torque control signal less than maximum torque (0.236 N.m), especially when external disturbances are applied and navigating through static obstacles. Results show that the five steps-ahead prediction algorithm has better performance compared to one step-ahead for all the control methodologies because of a more complex control structure and taking into account future values of the desired one, not only the current value, as with one step-ahead method. The mean-square error method is used for each component of the state error vector to compare between each of the performance control methodologies in order to give better control results.

This thesis focuses on the problem of "autonomous agents". It is assumed that such agents want to be in a desired state which can be assessed by the agent itself when it observes the consequences of its own actions. Therefore the feedback from the motor output via the environment to the sensor input is an essential component of such a system. As a consequence an agent is defined in this thesis as a self-referential system which operates within a closed sensor- mot or-sensor feedback loop. The generic situation is that the agent is always prone to unpredictable disturbances which arrive from the outside, i.e. from its environment. These disturbances cause a deviation from the desired state (for example the organism is attacked unexpectedly or the temperature in the environment changes, ...). The simplest mechanism for managing such disturbances in an organism is to employ a reflex loop which essentially establishes reactive behaviour. Reflex loops are directly related to closed loop feedback controllers. Thus, they are robust and they do not need a built-in model of the control situation. However, reflexes have one main disadvantage, namely that they always occur 'too late'; i.e., only after a (for example, unpleasant) reflex eliciting sensor event has occurred. This defines an objective problem for the organism. This thesis provides a solution to this problem which is called Isotropic Sequence Order (ISO-) learning. The problem is solved by correlating the primary reflex and a predictive sensor input: the result is that the system learns the temporal relation between the primary reflex and the earlier sensor input and creates a new predictive reflex. This (new) predictive reflex does not have the disadvantage of the primary reflex, namely of always being too late. As a consequence the agent is able to maintain its desired input-state all the time. In terms of engineering this means that ISO learning solves the inverse controller problem for the reflex, which is mathematically proven in this thesis. Summarising, this means that the organism starts as a reactive system and learning turns the system into a pro-active system. It will be demonstrated by a real robot experiment that ISO learning can successfully learn to solve the classical obstacle avoidance task without external intervention (like rewards). In this experiment the robot has to correlate a reflex (retraction after collision) with signals of range finders (turn before the collision). After successful learning the robot generates a turning reaction before it bumps into an obstacle. Additionally it will be shown that the learning goal of 'reflex avoidance' can also, paradoxically, be used to solve an attraction task.

This thesis is about appearance-based topological mapping for mobile robots using vision and laser. Our goal is life-long continual operation in outdoor unstruc- tured workspaces. We present a new probabilistic framework for appearance-based mapping and navigation incorporating spatial and visual appearance. Locations are encoded prob- abilistically as random graphs possessing latent distributions over visual features and pair-wise euclidean distances generating observations modeled as 3D constellations of features observed via noisy range and visual detectors. Multi-modal distributions over inter-feature distances are learnt using non-parametric kernel density estima- tion. Inference is accelerated by executing a Delaunay tessellation of the observed graph with minimal loss in performance, scaling log-linearly with scene complexity. Next, we demonstrate how a robot can, through introspection and then targeted data retrieval, improve its own place recognition performance. We introduce the idea of a dynamic sampling set, the onboard workspace representation, that adapts with increasing visual experience of continually operating robot. Based on a topic based probabilistic model of images, we use a measure of perplexity to evaluate how well a working set of background images explains the robot’s online view of the world. O/ine, the robot then searches an external resource to seek additional background images that bolster its ability to localize in its environment when used next. Finally, we present an online and incremental approach allowing an exploring robot to generate apt and compact summaries of its life experience using canon- ical images that capture the essence of the robot’s visual experience-illustrating both what was ordinary and what was extraordinary. Leveraging probabilistic topic models and an incremental graph clustering technique we present an algorithm that scales well with time and variation of experience, generating a summary that evolves incrementally with the novelty of data.

In urban search and rescue (USAR) operations, there is a considerable amount of danger faced by rescuers. The use of mobile robots can alleviate this issue. Coordinating the search effort is made more difficult by the communication issues typically faced in these environments, such that communication is often restricted. With small numbers of robots, it is necessary to break communication links in order to explore the entire environment. The robots can be viewed as a broken ad hoc network, relying on opportunistic contact in order to share data. In order to minimise overheads when exchanging data, a novel algorithm for data exchange has been created which maintains the propagation speed of flooding while reducing overheads. Since the rescue workers outside of the structure need to know the location of any victims, the task of finding their locations is two parted: 1) to locate the victims (Search Time), and 2) to get this data outside the structure (Delay Time). Communication with the outside is assumed to be performed by a static robot designated as the Command Station. Since it is unlikely that there will be sufficient robots to provide full communications coverage of the area, robots that discover victims are faced with the difficult decision of whether they should continue searching or return with the victim data. We investigate a variety of search techniques and see how the application of biological foraging models can help to streamline the search process, while we have also implemented an opportunistic network to ensure that data are shared whenever robots come within line of sight of each other or the Command Station. We examine this trade-off between performing a search and communicating the results.

This thesis is about the intelligent compression of large 3D point cloud datasets. The non-parametric method that we describe simultaneously generates a continuous representation of the workspace surfaces from discrete laser samples and decimates the dataset, retaining only locally salient samples. Our framework attains decimation factors in excess of two orders of magnitude without significant degradation in fidelity. The work presented here has a specific focus on gathering and processing laser measurements taken from a moving platform in outdoor workspaces. We introduce a somewhat unusual parameterisation of the problem and look to Gaussian Processes as the fundamental machinery in our processing pipeline. Our system compresses laser data in a fashion that is naturally sympathetic to the underlying structure and complexity of the workspace. In geometrically complex areas, compression is lower than that in geometrically bland areas. We focus on this property in detail and it leads us well beyond a simple application of non-parametric techniques. Indeed, towards the end of the thesis we develop a non-stationary GP framework whereby our regression model adapts to the local workspace complexity. Throughout we construct our algorithms so that they may be efficiently implemented. In addition, we present a detailed analysis of the proposed system and investigate model parameters, metric errors and data compression rates. Finally, we note that this work is predicated on a substantial amount of robotics engineering which has allowed us to produce a high quality, peer reviewed, dataset - the first of its kind.

Les microrobots magnétiques, qui nagent en utilisant des modes de propulsion bio-inspirées, apparaissent très prometteurs pour la manipulation et la caractérisation d'objets à l'échelle microscopique dans des environnements confinés et très restreints, contrairement aux méthodes de micromanipulation classiques. La littérature propose une variété de microrobots avec des formes géométriques et des propriétés magnétiques différentes. Les commandes en mouvement proposées restent cependant simples, peu précises et insuffisamment robustes pour la réalisation de tâches réelles. De plus, il subsiste une incertitude sur le fait que tous ces micronageurs artificiels peuvent accomplir les mêmes tâches avec une performance égale. L'objectif de cette thèse consiste alors à proposer : des commandes de mouvement génériques par asservissement visuel dans l'espace pour tous les types de micronageurs avec des contraintes non holonomes afin d'améliorer les performances de ces micronageurs, un ensemble de critères de comparaison entre des robots avec une topologie ou un mode de propulsion différents pour le choix du micronageur le plus performant pour réaliser une tâche particulière. Des lois de commande de suivi de chemin dans l'espace sont synthétisées et validées expérimentalement sur des nageurs hélicoïdal et flexible sous différentes conditions. Ces robots évoluent dans un fluide à faible nombre de Reynolds, imitant respectivement le mécanisme de locomotion des bactéries et des spermatozoïdes et sont actionnés par un champ magnétique uniforme. Ces deux classes de nageurs possèdent une géométrie et un mode d'actionnement différents. Leurs performances sont ainsi comparées. / Magnetic microrobots, which swim using bio-inspired propulsion modes, appear very promising for manipulation and characterization of objects at microscopic scale inside confined and very restricted environments, unlike conventional micromanipulation methods. The literature proposes a variety of microrobots with different geometric shapes and magnetic properties. However, the motion controls proposed remain simple, imprecise and insufficiently robust for performing real tasks. In addition, there is still uncertainty that all these artificial microswimmers can accomplish the same tasks with equal performance. The objective of this thesis is thus to propose : generic motion controls by visual servoing in space for all kinds of microswimmers with nonholonomic constraints in order to improve the microswimmer performances, a set of comparison criteria between robots with a different topology or propulsion mode for choosing the most efficient microswimmer in order to perform a specific task. Path following control laws in space are synthesized and experimentally validated on helical and flexible swimmers under different conditions. These robots operate in low Reynolds number fluid, imitating respectively bacteria and spermatozoa and are actuated with uniform magnetic field. These two classes of swimmers have different actuation mode and geometric shape. Their performances are thus compared according to the task to be performed, the environment in which the robots evolve and the manufacturing constraints.