Download files
Abstract
Autonomous Ground Vehicles (AGVs) are increasingly being used in more complex, harsh and remote environments because of their ability to replace a human driver, removing them from danger. However, the environments that are being targeted for the application of AGVs frequently occur in areas that have unstructured, uneven and non-uniform terrain. Safe and accurate navigation through such environments requires reliable terrain perception methods capable of identifying friction characteristics and an accurate system model, both are required in order to reduce the system uncertainties. Currently, methods for reducing these uncertainties are not well established.
This thesis researched the sources of uncertainty that an AGV is subject to whilst moving through an unstructured and non-uniform terrain. A number of key sources of uncertainty were identified, with changes in terrain type and thus the coefficient of friction being one key component, as well as the dynamic behaviour of the tyre interacting with the terrain, especially during cornering maneuvers.
Furthermore, it develops a novel terrain perception method which utilises purely non-semantic spatial information. The Extended Range Texture Analysis (ERTA) method is capable of providing an accurate terrain model that accounts for the identified terrain frictional and geometric characteristics, which reduces the uncertainties in terrain frictional characteristics that exists in the system model. The ERTA method can provide a solution for classifying the terrain types in an environment even in the absence of visual or semantic spatial features in a scene.
Two novel tyre models were also developed in this thesis: a simplified Friction Dependant tyre model, and a 3D Analytical Dynamic tyre model. The purpose of developing these models is to enable the incorporation of the sources of uncertainties, including both frictional characteristics provided by the terrain model and tyre dynamic effects. The simulation results of both the developed tyre models were compared against the widely accepted experimentally derived Magic Formula tyre model.
Following on from this research, control system designers can incorporate the developed tyre models and perception method to reduce the uncertainties in the system parameters and disturbances to design significantly improved controllers.