Development of behavioural strategies for indoor mobile navigation has become a challenging and practical issue in a cluttered indoor environment, such as a hospital or factory, where there are many static and moving objects, including humans and other robots, all of which trying to complete their own specific tasks; some objects may be moving in a similar direction to the robot, whereas others may be moving in the opposite direction. The key requirement for any mobile robot is to avoid colliding with any object which may prevent it from reaching its goal, or as a consequence bring harm to any individual within its workspace. This challenge is further complicated by unobserved objects suddenly appearing in the robots path, particularly when the robot crosses a corridor or an open doorway. Therefore the mobile robot must be able to anticipate such scenarios and manoeuvre quickly to avoid collisions. In this project, a hybrid control architecture has been designed to navigate within dynamic environments. The control system includes three levels namely: deliberative, intermediate and reactive, which work together to achieve short, fast and safe navigation. The deliberative level creates a short and safe path from the current position of the mobile robot to its goal using the wavefront algorithm, estimates the current location of the mobile robot, and extracts the region from which unobserved objects may appear. The intermediate level links the deliberative level and the reactive level, that includes several behaviours for implementing the global path in such a way to avoid any collision. In avoiding dynamic obstacles, the controller has to identify and extract obstacles from the sensor data, estimate their speeds, and then regular its speed and direction to minimize the collision risk and maximize the speed to the goal. The velocity obstacle approach (VO) is considered an easy and simple method for avoiding dynamic obstacles, whilst the collision cone principle is used to detect the collision situation between two circular-shaped objects. However the VO approach has two challenges when applied in indoor environments. The first challenge is extraction of collision cones of non-circular objects from sensor data, in which applying fitting circle methods generally produces large and inaccurate collision cones especially for line-shaped obstacle such as walls. The second challenge is that the mobile robot cannot sometimes move to its goal because all its velocities to the goal are located within collision cones. In this project, a method has been demonstrated to extract the colliii sion cones of circular and non-circular objects using a laser sensor, where the obstacle size and the collision time are considered to weigh the robot velocities. In addition the principle of the virtual obstacle was proposed to minimize the collision risk with unobserved moving obstacles. The simulation and experiments using the proposed control system on a Pioneer mobile robot showed that the mobile robot can successfully avoid static and dynamic obstacles. Furthermore the mobile robot was able to reach its target within an indoor environment without causing any collision or missing the target.

Accurate control techniques for pinpoint planetary landing - i.e., the goal of achieving landing errors in the order of 100m for unmanned missions - is a complex problem that have been tackled in different ways in the available literature. Among other challenges, this kind of control is also affected by the well known trade-off in UAV control that for complex underlying models the control is sub-optimal, while optimal control is applied to simplifed models. The goal of this research has been the development new control algorithms that would be able to tackle these challenges and the result are two novel optimal control algorithms namely: OQTAL and HEX2OQTAL. These controllers share three key properties that are thoroughly proven and shown in this thesis; stability, accuracy and adaptability. Stability is rigorously demonstrated for both controllers. Accuracy is shown in results of comparing these novel controllers with other industry standard algorithms in several different scenarios: there is a gain in accuracy of at least 15% for each controller, and in many cases much more than that. A new tuning algorithm based on swarm heuristics optimisation was developed as well as part of this research in order to tune in an online manner the standard Proportional-Integral-Derivative (PID) controllers used for benchmarking. Finally, adaptability of these controllers can be seen as a combination of four elements: mathematical model extensibility, cost matrices tuning, reduced computation time required and finally no prior knowledge of the navigation or guidance strategies needed. Further simulations in real planetary landing trajectories has shown that these controllers have the capacity of achieving landing errors in the order of pinpoint landing requirements, making them not only very precise UAV controllers, but also potential candidates for pinpoint landing unmanned missions.

No description available.

Robots have been widely used for various applications, such as smart transportation, environment monitoring, surveillance, search and rescue. Autonomous navigation, as a core prerequisite for the robots to successfully realise these applications, relies heavily on robot localisation. Global Positioning System (GPS) fails to satisfy many applications in robotics in terms of accuracy and availability. Therefore, robot localisation in GPS-denied 3D space is in great demand. However, due to sensor noise and real world uncertainty, robot localisation in GPS-denied 3D space is a challenging problem. The work in this thesis describes three novel localisation algorithms to localise the robots accurately and efficiently in different .scenarios. .optimal state estimation, including filter based and optimisation based methods, is adopted to elegantly deal with the noise and the uncertainty in a probabilistic perspective. Firstly, a Moving Horizon Estimation (MHE) based localisation algorithm is proposed for single beacon based robot localisation. The performance and observability analyses are also conducted to evaluate the proposed method. Secondly, single beacon based multi-robot cooperative localisation problem is addressed by a constrained MHE based approach. Its discussion answers why and how multi-robot cooperation and optimisation constraints benefit the localisation system. The initial pose estimation problem and observability analysis of the multi-robot system are also studied. Thirdly, an unscented Kalman filter based algorithm is proposed for Vision-aided Inertial Navigation System (VINS) by only using low-cost camera and Inertial Measurement Unit (IMU) to perform pose estimation and camera-IMU extrinsic self-calibration. Trifocal tensor based geometric constraints and point transfer of three-view geometry are incorporated into VINS. Tested by both simulations and experiments, the proposed methods are verified to be effective for robot localisation in GPS-denied 3D space.

No description available.

No description available.

With current robotic technologies, it generally remains unreliable to use fully autonomous robots in high-risk robotic applications such as search and rescue, surveillance or exploration in disaster scenarios. One of the main issues comes from the fact that unstructured real-world environments are dynamic and full of interventions. Therefore, for autonomous robots to operate in such environments, the ability to adapt to both internal and external environmental changes is crucial. Being unable to deal with such changes not only could downgrade the performance of the robots but also potentially cause devastating consequences in risky environments. Looking towards nature, it can be observed that biological organisms can cope well with the dynamic unpredictability of real-world environments. One of the key properties which assist biological organisms is the ability to adapt to changing environments by the utilization of hormones in response to environmental cues. This biological feature provides an inspiration for this research which investigates a novel Artificial Hormone Network architecture in providing adaptability for autonomous robots to deal with both internal and external environmental changes in simulations of unstructured real-world environments. The Artificial Hormone Network architecture proposes a new method which allows constructions and interactions of several hormones in order to provide adaptability for autonomous robots in different application scenarios. Two Artificial Hormone Networks (AHN1 and AHN2) are proposed and investigated in this research. Results from experiments correspondingly report better performance in dealing with considered internal and external environmental changes on a robot implemented with the Artificial Hormone Networks than a robot implemented without them. Another important aspect of the Artificial Hormone Network architecture is the ability to be constructed automatically to provide particular adaptability using Cartesian Genetic Programming. Experiment results show that the construction of Artificial Hormone Networks can be evolved and that this evolved system not only performed to a level of adaptability that was acceptable but actually performed better than the “hand-coded” system.

In complex systems, it has been observed that the parts within the system are "loosely coupled". Loose coupling means that the parts of the system interact in some way, and as long as this interaction is maintained the parts can evolve independently. Detrimental evolutionary changes within one part of the system do not negatively affect other parts. Overall system functionality is maintained, leading to faster evolution. In swarm robotic systems there are multiple robots working together to achieve a shared goal. However it is not always obvious how to program the actions of the robots such that the desired aggregate behaviour emerges. One solution is to use a genetic algorithm to evolve robot controllers, this approach is called "Evolutionary Swarm Robotics". This thesis makes the case that swarm robotic systems are complex systems, and hypothesises that loose coupling between the robots in a swarm would lead to faster evolution. Robot swarms are investigated where robots describe environmental features to each other as part of a foraging task. Multiple descriptions can be used to describe a feature. The mappings between feature descriptions, and the signals used to express those descriptions, are manipulated. By doing this, the interactions between robots can change over time or stay the same. Results show that loose coupling leads to higher swarmfitnesses because it makes the communicated information easier to interpret. However there are some subtleties in its working. We also observe that if some of the information is not useful for completing the task, this negatively affects swarm fitness regardless of coupling. This problem can be mitigated by using loose coupling. This research has implications for the design of communication within robot swarms. Before evolution, it is difficult to know what information is relevant. This research shows that sharing unnecessary information between robots is detrimental to swarm fitness because the cost of interpreting information can be greater than the benefit gained from the information. Loose coupling can reduce, but not eliminate, the evolutionary cost of interpreting multiple pieces of information in exchange for slower message transmission.

In the thesis we solve the shared-control problem for three classes of systems: a class of linear mechanical systems, mobile robots and rear wheel drive cars, via full state feedback or output feedback while ensuring that all the state constraints on the closed-loop systems are satisfied. To design the feedback controller for a system with state constraints we firstly remove all the constraints by changing the coordinates through a logarithmic function. Then the back-stepping method is used to design the controller and a Lyapunov-like analysis is used to prove stability properties of the closed-loop system. The shared-control algorithm is based on a hysteresis switch which reduces oscillations when changing the control authority from the human operator to the feedback controller or vice-versa. Unlike other shared-control methods, formal properties of the closed-loop systems with the shared-control have been rigorously established. We start the design of the full state-feedback shared-controller with the assumption that the admissible Cartesian configuration set Pa of the system is a time-invariant convex set defined by a group of linear inequalities. Then the results are extended to the design of shared-controllers via output feedback. In the cases in which only output feedback is available, we can solve the problem by either developing an observer or 'remodeling' the system. Through system remodeling we are able to deal with any shape of the admissible configuration set Pa, even time-varying ones. Simulation results help to illustrate how the shared-controller works and show its effectiveness. The state of the closed-loop system with the shared-control never violates the constraints. Experiments done on a mobile robot also demonstrate that the shared-control algorithm works well in practice and meets all safety requirements. In addition, the experimental results match the simulation ones, indicating that the modeling approximations are reasonable and suitable.

This thesis is concerned with the task of efficiently recognising the particular instance of a scene depicted in a query image, with applications in robot navigation including loop closure, global localisation and topological navigation. Three novel frameworks are proposed, each based on learning scene models by tracking local features to form sets of landmarks. Recognition then proceeds by considering 2D constraints between pairs of local feature correspondences to efficiently approximate global scene geometry. First, the inter-image and intra-image pairwise geometries are considered to reduce feature correspondences to a more succinct set for a RANSAC-based 3D geometry constraint. A Hough-transform voting scheme based on inter-image correspondences weakly prunes the set of correspondences, after which intra-image geometries constrain the relative image positions of correspondences to eliminate unrealistic configurations. This idea is first proposed in an image retrieval application, and then extended to scene recognition where relative landmark positions are learned explicitly per scene. Second, a method is introduced to embed 2D pairwise geometry directly in an inverted index, to allow for fast scene recognition without 3D estimations. A set of discrete geometric words are extracted for a query image, and passed through the index to find examples of such pairwise configurations in the database. A global geometry constraint is then proposed by considering a maximum-clique approach to an adjacency matrix of correspondences. Third, a global topological localisation system is investigated which learns a naive Bayesian network for each landmark, to efficiently approximate global geometry without a fully-connected model. Long-term robot navigation is then addressed by learning scene models in an incremental manner, and updating the dynamic properties of landmarks accordingly. Experiments were performed on a new challenging dataset obtained by manually walking along a 7km path in a park and urban district, to capture long-term effects over an 8 month period.