Marvin K. Bugeja

Dual Adaptive Dynamic Control of Mobile Robots Using Neural Networks

This paper proposes two novel dual adaptive neural control schemes for the dynamic control of nonholonomic mobile robots. The two schemes are developed in discrete time, and the robots nonlinear dynamic functions are assumed to be unknown. Gaussian radial basis function and sigmoidal multilayer perceptron neural networks are used for function approximation. In each scheme, the unknown network parameters are estimated stochastically in real time, and no preliminary offline neural network training is used. In contrast to other adaptive techniques hitherto proposed in the literature on mobile robots, the dual control laws presented in this paper do not rely on the heuristic certainty equivalence property but account for the uncertainty in the estimates. This results in a major improvement in tracking performance, despite the plant uncertainty and unmodeled dynamics. Monte Carlo simulation and statistical hypothesis testing are used to illustrate the effectiveness of the two proposed stochastic controllers as applied to the trajectory-tracking problem of a differentially driven wheeled mobile robot.

Double Exponential Smoothing for predictive vision based target tracking of a wheeled mobile robot

This paper describes the design of a novel nonlinear kinematic controller which allows a wheeled mobile robot to track a moving target at a given separation distance. The Double Exponential Smoothing algorithm is employed to deal with uncertainties in the measurements and to acquire a predictive estimate for the robots relative position. This estimate is used to automatically adjust the proportional gain of the controller in order to regulate the tracking error.

Multilayer Perceptron Adaptive Dynamic Control for Trajectory Tracking of Mobile Robots

This paper presents a novel functional-adaptive dynamic controller for trajectory tracking of nonholonomic wheeled mobile robots. The controller is developed in discrete-time and employs a multilayer perceptron neural network for the estimation of the robots nonlinear dynamic functions, which are assumed to be completely unknown. On-line weight tuning is achieved by employing the extended Kalman filter algorithm, based on a specifically formulated stochastic inverse dynamic identification model of the mobile base. A discrete-time dynamic control law employing the estimated functions is proposed and cascaded with a trajectory tracking kinematic controller. The performance of the complete system is analysed and compared by realistic simulations

Dual adaptive extremum control of Hammerstein models

This paper proposes a dual adaptive control methodology for extremum control of stochastic Hammerstein systems that are subject to uncertainty and characterised by a nonlinear second-order polynomial function. The design is based on an explicit-type innovations dual cost function that leads to a control law which balances out the need for caution when using uncertain parameter estimates to effect control actions, with the conflicting need to probe the system input so as to reduce parameter uncertainty. This extremum control problem is more challenging than conventional adaptive systems because the reference input is itself a non-linear function of the unknown system parameters. The controllers performance is analysed through extensive Monte Carlo simulations and shown to be superior to other types of adaptive control systems that use a certainty equivalence assumption.

Multilayer Perceptron Adaptive Dynamic Control of Mobile Robots: Experimental Validation

This paper presents experimental results acquired from the implementation of an adaptive control scheme for nonholonomic mobile robots, which was recently proposed by the same authors and tested only by simulations. The control system comprises a trajectory tracking kinematic controller, which generates the reference wheel velocities, and a cascade dynamic controller, which estimates the robot’s uncertain nonlinear dynamic functions in real-time via a multilayer perceptron neural network. In this manner precise velocity tracking is attained, even in the presence of unknown and/or time-varying dynamics. The experimental mobile robot, designed and built for the purpose of this research, is also presented in this paper.

A Comparison of Two Pressure Control Concepts for Hydraulic Offshore Wind Turbines

Current research in offshore wind turbines is proposing a novel concept of using seawater-based hydraulics for large-scale power transmission and centralized electrical generation. The objective of this paper is to investigate the control of such an open-loop circuit, where a fixed line pressure is desirable for the sake of efficiency and stability. Pressure control of the open-loop hydraulic circuit presents an interesting control challenge due to the highly fluctuating flow rate along with the nonlinear behavior of the variable-area orifice used by the pressure controller. The present analysis is limited to a single turbine and an open-loop hydraulic line with a variable-area orifice at the end. A controller is proposed which uses a combination of feed-forward compensation for the nonlinear part along with a feedback loop for correcting any errors resulting from inaccuracies in the compensator model. A numerical model of the system under investigation is developed in order to observe the behavior of the controller and the advantages of including the feedback loop. An in-depth analysis is undertaken, including a sensitivity study of the compensator accuracy and a parametric analysis of the actuator response time. Finally, a Monte Carlo analysis was carried out in order to rank the proposed controller in comparison to a simple feed-forward controller and a theoretical optimally tuned controller. Results indicate an advantageous performance of the proposed method of feedback with feed-forward compensation, particularly its ability to maintain a stable line pressure in the face of high parameter uncertainty over a wide range of operating conditions, even with a relatively slow actuation system.

Application of Sliding Mode Control to the Ball and Plate Problem

This paper proposes and investigates the application of sliding mode control to the ball and plate problem. The nonlinear properties of the ball and plate control system are first presented. Then the experimental setup designed and built specifically for the purpose of this research is discussed. The paper then focuses on the implementation and thorough evaluation of the experimental results obtained with two different control schemes: the linear full-state feedback controller and the sliding mode controller. The latter control strategy was selected for its robust and order reduction properties. Finally the control performance of the two controllers is analysed. The sliding controller manages to obtain a faster and more accurate operation for continuously changing reference inputs. The robustness of the proposed control scheme is also verified, since the systems performance is shown to be insensitive to parameter variations.

Dual Adaptive Neurocontrol of Mobile Robots Using the Unscented Transform: Monte Carlo and Experimental Validation

In contrast to most adaptive schemes, dual adaptive controllers do not rely on the heuristic certainty equivalence assumption, but aim to strike a balance between estimation and control at all times. Yet, few such controllers have ever been implemented and tested in practice, especially within the context of intelligent control, and to the best of our knowledge none on mobile robots. With the help of Mont Carlo simulation and real-life experiments, this article presents and validates a novel dual adaptive neurocontroller based on the unscented transform, for the dynamic control of nonholonomic wheeled mobile robots. The robot nonlinear dynamic functions are unknown to the controller and a multilayer perceptron neural network, trained via an unscented Kalman predictor, is used for their approximation in real-time. Moreover, the proposed novel dual adaptive control law employs the unscented transform to improve further the system’s performance.

Control of an Open-Loop Hydraulic Offshore Wind Turbine Using a Variable-Area Orifice

The viability of offshore wind turbines is presently affected by a number of technical issues pertaining to the gearbox and power electronic components. Current work is considering the possibility of replacing the generator, gearbox and electrical transmission with a hydraulic system. Efficiency of the hydraulic transmission is around 90% for the selected geometries, which is comparable to the 94% expected for conventional wind turbines. A rotor-driven pump pressurises seawater that is transmitted across a large pipeline to a centralised generator platform. Hydroelectric energy conversion takes place in Pelton turbine. However, unlike conventional hydro-energy plants, the head available at the nozzle entry is highly unsteady. Adequate active control at the nozzle is therefore crucial in maintaining a fixed line pressure and an optimum Pelton turbine operation at synchronous speed. This paper presents a novel control scheme that is based on the combination of proportional feedback control and feed forward compensation on a variable area nozzle. Transient domain simulation results are presented for a Pelton wheel supplied by sea water from an offshore wind turbine-driven pump across a 10 km pipeline.Copyright

Neuro-adaptive dynamic control for mobile robots: Experimental validation

This paper reports on the design and implementation of a neuro-adaptive controlled nonholonomic mobile robot. It presents experimental results to validate the employed control scheme on a physical setup for the first time, after it was originally proposed by the same authors and tested by simulations only. The control system is composed of a trajectory tracking kinematic controller which generates the reference wheel velocities, and a cascaded dynamic controller which employs a neural network for the real-time estimation of the robots nonlinear dynamics so as to attain precise velocity tracking, even in the presence of unknown and/or time-varying dynamics. Details about the hardware and software setup, as well as salient implementation issues are also reported in this work.