Obstacles in the path of travel are a common feature of natural and built environments. The ability to negotiate such terrain, traversing rather than avoiding obstacles, increases the versatility of animal locomotion. The aim of this thesis was to investigate how and why humans and dogs select characteristic changes to gait for obstacle traversal. In a series of experiments, subjects were presented with different types of obstacle and the kinematics of locomotion used to traverse them recorded. Fundamentally different traversal strategies viable for overlapping ranges of obstacle dimensions were identified based on the characteristic kinematics of traversal behaviour. Obstacle traversal strategies were systematically selected based on the dimensions of the obstacle. Adaptations to gait in advance of traversal were shown generally to be localised to the obstacle. The maximum amount by which step length was increased to leap over an obstacle was invariant with approach speed, although the mechanism for lengthening the step changed fundamentally with increasing speed of travel. Preferred traversal dynamics were associated with the minimisation of mechanical energy cost in some but not all of the obstacle conditions investigated. The position of the baseline of the obstacle was identified as an important property for the visual judgement of obstacle position. This body of work provides insight into the control of legged locomotion over irregular terrain. It has potential applications in the design of autonomous legged vehicles.

This thesis presents novel methods for agent-based mission planning sys- tem for multiple unmanned autonomous vehicles (MUAV). MUAV mission planning is the pinnacle of the intelligent systems. The ever-changing ad- verse environment and real-time co-operative decision making adds to the complexity of the system. Due to these inherent complexities, the progress from autonomic to autonomous MUAV system is still in its infancy. The conventional methods of distributed planning and decision making have shown some benefits but are not sufficient to develop the co-operative intelligence in MUAV system. Here we present agent-based approach for developing MUAV mission planning system by in-cooperative intelligent behaviours for self-organisation, self-awareness and intelligent decision making. These co-operative behaviours are aimed to add autonomy to the system. The requirements, interactions, functionalities and the role of these methods in overall system are established by in-depth study of existing control frameworks for MUAV management system. We also present a unified framework for NIUAV mission planning. This functional based framework provides a better yet simple understanding of the otherwise complex system. It serves the purpose of providing a better understanding of the challenges and opportunities in development of MUAV system by providing logical system construction of the MUAV mission planning in detail. To facilitate the self-awareness of the MUAV system, an efficient Advanced Integrated Method (AIM) path planning method is developed. AIM generates optimum obstacle free path from source to destination the consideration of UAV's safety. It combines the existing methods of Artificial potential, Maximum Clearance and Self Organizing Maps (SOM) for guaranteed convergence. We also present a model for the prediction of the future states of moving targets as stochastic processes with associated learned transition probabilities using Discrete Markov Chains (DMC). These predictions are then used for developing interception based target tracking. These predictions also provide fair and effective mean for target allocation among multiple UAV's and for target selection in the presence of multiple targets. The co-operative behaviour for MUAV system is further supported by a new and effective method of self-organisation. Inspired from thermodynamic systems, it introduces co-operative self-allocation of mission space with the objective of sharing surveillance responsibilities. These methods for path planning, prediction based decision making and self-allocation collectively provide the groundwork for building autonomous MUAV system.

This thesis presents investigations into development of modelling and control of flexible structures using swarm intelligence optimisation techniques. A smart flexible beam structure is used in this work as a candidate application. The smart flexible beam model is developed using finite difference method and a methodology of incorporating piezoelectric patch actuator into finite difference model is presented. The simulation model is developed in MATLAB/SIMULINK environment as a platform for test and verification of the control approaches developed in this work. Many heuristic search algorithms have been inspired by nature such as genetic algorithm (natural evolution), artificial neural network (biological neuron) and artificial immune system (immune system) where the algorithms try to mimic the biological process. Addition to nature inspired algorithms is the swarm intelligence method which has been inspired by the natural behaviour of a group of insects like foraging, flocking and schooling in ants, bees, fish and birds where particle swam optimisation (PSO) and ant colony optimisation (ACO) are the most popular methods. The study of parameter setting for PSO and continuous ACO (ACOr) is studied through parametric modelling of the beam. The performance of each algorithm in terms of computational time and convergence is discussed. In this study, vibration control of a flexible beam structure is developed based on the principle of wave interference, to result in optimal cancellation with adaptive model-based control and adaptive direct control. A single objective optimisation algorithm is developed and implemented using PSO and continuous ACO considering two conditions; optimisation of controller with pre-selected location of sensor and actuator and simultaneous optimisation of controller parameters and sensor and actuator location in single-input-single-output and single-input-multiple-output configurations. While single objective optimisation provides only one solution, the use of multi-objective optimisation results in several solutions to choose for implementation. An approach of multi-objective optimisation of controllers' parameters and sensor/actuator location is developed based on minimising vibration energy and minimising actuator force. Multi-objective PSO and multi-objective ACOr algorithms are developed in finding optimal system setup and controller parameters for AYC of the beam. Both PSO and ACO based algorithms are tested and their performances assessed in vibration control of the beam.

No description available.

No description available.

The work reported in this thesis aims to design and develop a new neuro-fuzzy control system for robotic manipulators using Machine Learning Techniques, Fuzzy Logic Controllers, and Fuzzy Neural Networks. The main idea is to integrate these intelligent techniques to develop an adaptive position controller for robotic manipulators. This will finally lead to utilising one or two coordinated manipulators to perform upper-limb rehabilitation. The main target is to benefit from these intelligent techniques in a systematic way that leads to an efficient control and coordination system. The suggested control system possesses self-learning features so that it can maintain acceptable performance in the presence of uncertain loads. Simulation and modelling stages were performed using dynamical virtual reality programs to demonstrate the ideas of the control and coordination techniques. The first part of the thesis focuses on the development of neuro-fuzzy models that meet the above requirement of mimicking both kinematics and dynamics behaviour of the manipulator. For this purpose, an initial stage for data collection from the motion of the manipulator along random trajectories was performed. These data were then compacted with the help of inductive learning techniques into two sets of if-then rules that form approximation for both of the inverse kinematics and inverse dynamics of the manipulator. These rules were then used in fuzzy neural networks with differentiation characteristics to achieve online tuning of the network adjustable parameters. The second part of the thesis introduces the proposed adaptive neuro-fuzzy joint-based controller. To achieve this target, a feedback Fuzzy-Proportional-Integral-Derivative incremental controller was developed. This controller was then applied as a joint servo-controller for each robot link in addition to the main neuro-fuzzy feedforward controller used to compensate for the dynamics interactions between robot links. A feedback error learning scheme was applied to tune the feedforward neuro-fuzzy controller online using the error back-propagation algorithm. The third part of the thesis presents a neuro-fuzzy Cartesian internal model control system for robotic manipulators. The neuro-fuzzy inverse kinematics model of the manipulator was used in addition to the joint-based controller proposed and the forward mathematical model of the manipulator in an adaptive internal model controller structure. Feedback-error learning scheme was extended to tune both of the joint-based neuro-fuzzy controller and the neuro-fuzzy internal model controller online. The fourth part of the thesis suggests a simple fuzzy hysteresis coordination scheme for two position-controlled robot manipulators. The coordination scheme is based on maintaining certain kinematic relationships between the two manipulators using reference motion synchronisation without explicitly involving the hybrid position/force control or modifying the existing controller structure for either of the manipulators. The key to the success of the new method is to ensure that each manipulator is capable of tracking its own desired trajectory using its own position controller, while synchronizing its motion with the other manipulator motion so that the differential position error between the two manipulators is reduced to zero or kept within acceptable limits. A simplified test-bench emulating upper-limb rehabilitation was used to test the proposed coordination technique experimentally.

This work proposes a biologically inspired collective behaviour for a team of co-operating robots. Collective behaviour is achieved by controlling the local interactions among a set of identical mobile robots, each robot performing a set of simple behaviours in order to realise group goals. A modification of the subsumption architecture is proposed for implementing control of individual robots. This architecture is adopted because it is computationally inexpensive and potentially suitable for low-level reactive and reflexive behaviours. In this scenario, the individual behaviours of the robots have different aims, which may cause conflict. To address this issue, a fuzzy logic-based approach for multiple behaviour coordination within each robot is proposed. The work also focuses on the development of intelligent multi-agent robot teams capable of acting autonomously and of collaborating in a dynamic environment to achieve team objectives. A knowledge-based software architecture is proposed that enables these robots to select co-operative behaviours and to adapt their performance during the specified time of the mission. These abilities are important because of uncertainties in the environmental conditions and because of possible functional failures in some team members. Improvement in team performance is achieved by updating the control of the robots based on knowledge acquired on-line. This architecture is implemented in a simulated team of mobile robots performing a proof-of-concept collaborative task. The results show a significant improvement in overall group performance and the robot team is able to achieve adaptive cooperative control despite dynamic changes in the environment and variation in the capabilities of the team members. Finally, a task involving real mobile robots is undertaken to demonstrate a practical, though simplified, implementation of the proposed collective behaviour.

This thesis focuses on visual servoing (VS) control systems, particularly on image-based visual servoing (IBVS) control structures. In IBVS, the error signal is computed in the image plane and the regulation commands are generated with respect to such error by means of a visual Jacobian. The main design challenge is the high latency of the visual sensor which affects the overall performance and limits the design. The primary objective is to develop a complete framework for simulation and real-time experimentation of VS schemes. One commercial CCD camera is attached to the TQ MA2000 robotic manipulator. The framework has been employed to investigate the use of RL algorithms to increase the performance of the IBVS control structure. The classic RL actor-critic structure has been used to perform on-line adjustment of the gains driving the linear trajectory regulator inside the IBVS control structure. The neural system learns directly from data in the image space and the state of the robot. Two feedforward networks are used, the actor directly modifies the regulator gains whereas the adaptive critic stores and assigns action values. By using the adaptive heuristic critic approach (AHC), the training aims to achieve real-time improvement and adaptation without losing an acceptable regulation of the visual servoing task. A compact model and a flexible framework host the reinforcement learning algorithm in order to enable its inclusion within the IBVS control structure. The approach in this thesis has solved critical neuro-dynamic problems which are derived from the interaction between the imaging model and the robot’s dynamics. The VS toolkit also provides a real-time library to implement and test the IBVS control structure. The libraries have proven effective to construct both the linear IBVS and the RL-supported IBVS system thanks to its layered architecture which facilitates the inclusion of control en› tities of different nature such as the neural networks and the learning framework. Two case studies demonstrate the applicability of the CSC VS toolkit to integrate all the required components and to implement each of the VS experiments in real-time. Performance comparison between the linear IBVS and the RL-supported system are also documented to show the effectiveness of the actor-critic structure.

To enable personalised assistance, assistive robots benefit from building a user-specific model, so that the assistance is customised to the particular set of user abilities. Among various tasks in home environments, assistive dressing, which is greatly beneficial to people with upper-body movement limitations, remains a challenging task for humanoid robots. In this thesis, we aim to design, implement, and evaluate user modelling methods which can enable humanoid robots to provide personalised dressing assistance. We begin by proposing a user modelling method using vision information. We use Gaussian mixture models (GMMs) to model the movement space of the human upper-body joints to learn the reachable area of each joint. We enabled a Baxter humanoid robot to plan its dressing motion using the GMMs of the human joints and real-time pose estimation. The dressing assistance is personalised by fulfilling a reachability criterion. To compensate for the disadvantages of using vision information only, we proposed an online iterative path optimisation method based on adaptive moment estimation. We enabled the Baxter robot to search for the optimal personalised dressing path for human users using force information. The dressing assistance is personalised by fulfilling a comfort criterion. Finally, to enable personalised dressing assistance fulfilling both the reachability and the comfort criteria, we proposed a user modelling method using multi-modal information by combining the GMMs of the human upper-body joints with the online iterative path optimisation. Experiments on both the synthetic dataset and the real-world assistive dressing data showed that the proposed method can achieve a balance between the two criteria when searching for the optimal path.

No description available.