Search

The works reported in this thesis focus upon synthesising neural controllers for anthropomorphic robots that are able to manipulate objects through an automatic design process based on artificial evolution. The use of Evolutionary Robotics makes it possible to reduce the characteristics and parameters specified by the designer to a minimum, and the robot’s skills evolve as it interacts with the environment. The primary objective of these experiments is to investigate whether neural controllers that are regulating the state of the motors on the basis of the current and previously experienced sensors (i.e. without relying on an inverse model) can enable the robots to solve such complex tasks. Another objective of these experiments is to investigate whether the Evolutionary Robotics approach can be successfully applied to scenarios that are significantly more complex than those to which it is typically applied (in terms of the complexity of the robot’s morphology, the size of the neural controller, and the complexity of the task). The obtained results indicate that skills such as reaching, grasping, and discriminating among objects can be accomplished without the need to learn precise inverse internal models of the arm/hand structure. This would also support the hypothesis that the human central nervous system (cns) does necessarily have internal models of the limbs (not excluding the fact that it might possess such models for other purposes), but can act by shifting the equilibrium points/cycles of the underlying musculoskeletal system. Consequently, the resulting controllers of such fundamental skills would be less complex. Thus, the learning of more complex behaviours will be easier to design because the underlying controller of the arm/hand structure is less complex. Moreover, the obtained results also show how evolved robots exploit sensory-motor coordination in order to accomplish their tasks.

This thesis presents investigations into modelling and control of flexible manipulators in 3D motion. A finite difference (FD) simulation environment characterising a single-link manipulator in horizontal plane motion and corresponding experimental rig is considered as test bed in this work. The dynamic model of the system is derived using the Lagrange equation and discretised using the FD method. Mathematical models are developed based on physical laws and first principles to represent the dynamics of a physical system well. The simulation algorithm is extended by deriving mathematical model of a single-link flexible manipulator in vertical plane motion with inclusion of gravity. The dynamic behaviour of the system for vertical motion is formulated. The response of the manipulator is assessed in comparison to the corresponding theoretical ones in time and the frequency domains to verify the accuracy of the model in characterizing the behaviour of the flexible manipulator system. However, for systems with complex behaviour and non-linear dynamics, modelling based on first principles is often a formidable and undesirable task, so to solve. these problems, system identification proven as an excellent tool to model such complex systems, is adopted. The objective of system identification is to find exact or approximate model of a dynamic system based on observed system input and output. Linear and non-linear models for the flexible manipulator system are developed using system identification techniques. Linear parametric models, characterising the flexible manipulator system are obtained using the potential of recursive least squares (RLS) estimation, genetic algorithms (GAs) and particle swarm optimisation (PSO) techniques, and combination of estimation and optimisation methods with GARLS. Moreover, the development of PSO with spread factor (PSOSF) and momentum factor (PSOMF) are used to develop suitable model of the flexible manipulator. Furthermore, nonlinear models using multi-layer perceptron (MLP) neural networks (NNs) are developed. These comprise combination GA with inverse NN (GAINN) and PSO with inverse NN (PSOINN). Data from an experimental flexible manipulator system is used to model the system using auto regressive moving average (ARMA) model structure with one-step-ahead prediction.

Robotic manipulators have for years been used in place of humans where tasks are too dangerous, arduous or repetitive. Thanks to their extra degrees-of-freedom hyper- redundant manipulators offer an increased ability to move within physically constrained spaces; useful for inspection and maintenance operations in hard-to- access environments. However, being comprised of many links joined end-to-end, they usually exhibit complex behaviour, and as a result are difficult to model and implement in the real-world. Indeed, very few examples have been successfully deployed. This thesis considers the kinematics, dynamics, and control of these hyper- redundant manipulators. A form of inverse kinematics based on a Bezier curve was developed. The method reduced the complexity of the problem and removed the need to use Jacobian matrices. It compared well to other inverse kinematics methods, and also proved to be implementable on a real-world test-rig. The standard approach to controlling hyper-redundant manipulators imposes constraints on the mechanical design of hyper-redundant manipulators. Proportional- integral-derivative control necessitates the use of highly geared joints, and usually exacerbate problems with backlash. This thesis investigated the application of nonlinear controllers to hyper-redundant manipulators. Sliding mode control was shown to be superior, particularly in response to gravitational loads. Chattering was successfully suppressed via incorporation of a modified error-dependent switching gain - the control effort was reduced by 95%. These ideas were validated in preliminary tests on a real-world test-rig.

This thesis concerns a research project in which parallel kinematic manipulators consisting of different types of joints and varying number of limbs are assessed in terms of their output characteristics. To increase the orientation capability of a parallel kinematic machine, a small number of limbs is necessary lo reduce limb interference. An innovative algorithm is implemented based on physical limits of the joints and condition number calculation of inverse Jacobians. Larger workspaces are found in the cases where spherical joints are used on the lower end of the limb as compared to other joints. A three legged SPS system is selected for further ysis due to its minimum number of limbs and the use of spherical joints in order to gain maximum degree of freedom.

This research aims to add to the understanding of manipulation and actuator control with the objective of facilitating the future development of fully autonomous manipulating agents. As a first stage along this path, the work presents a simulated implementation of two techniques designed to control a two degree of freedom arm, equipped with six compliant actuators. After preliminary modelling with a static arm, the first technique, inspired by a biological motor theory called convergent force field control, seeks to use a multi-layer-perceptron (MLP) to steer a custom built dynamic simulation of the arm towards predefined targets within the workspace. The MLP is trained iteratively using mutation hill climbing. The work demonstrates that this approach is capable of creating arm responses that move towards a fixed target from any location within the workspace. That said, the learning approach is not reliable and often gets stuck in local minima. The section concludes by demonstrating the blending of two controllers, to create a smooth range of intermediate results. The second technique combines convergent force field control with another related biological control theory, called equilibrium trajectory control. Key to the success of this technique is the assumption that smooth natural movements of the tip of the arm can be generated by constant rate movement of the arm's equilibrium point, with respect to the workspace. The model proposed uses a form of convergent force field control to guide an internal representation of the equilibrium point towards a target at a constant rate. Various implementation options are considered with particular attention paid to the way the equilibrium point is encoded internally. Initial development of the trajectory generator and length encoder indicate that, for reasonable sizes of MLP, shoulder centred polar encoding is the most computationally efficient scheme. It is hoped that the work presented here will provide a foundation for techniques that will bridge the gap between behaviour based control and low level coordination of complex compliant articulated systems.