Terrain Characterization and Roughness Estimation for Simulation and Control of Unmanned Ground Vehicles
Type of Degreedissertation
MetadataShow full item record
This dissertation presents a methodology for generating artificial terrains for simulation of off-road vehicles. Furthermore it develops and evaluates methods for characterizing the terrain for the control of unmanned ground vehicles. The terrain is the principle source of chassis excitation in off-road vehicles and the control of the vehicle is dependent on effectively characterizing the terrain slope, roughness, and surface condition. The previous work in this area is presented and the areas for improvement are identified. The literature is vast and is categorized into works which have addressed various parts of the problems. It is advantageous for the development of autonomous vehicle systems to simulate the vehicle response over various terrains; this requires generating artificial terrains which are similar to real terrains. Two methods for generating terrains based on the Weierstrass-Mandelbrot (W-M) fractal function are presented. The generated surfaces are evaluated using the root mean squared elevation (RMSE) and power spectral density (PSD). A seven degree of freedom (7-DOF) suspension model is developed for the purpose of evaluating the response of the vehicle on the generated terrains. The vehicle response is used to introduce motion based metrics for characterizing the roughness of the terrain. The root mean squared (RMS) vertical acceleration, RMS roll rate, and RMS pitch rate are introduced as potential motion base metrics. Additionally the phase plane of various vehicle states is investigated as a means for understanding the vehicle state combined with the terrain roughness. A system for generating three dimensional point cloud maps of terrains is presented. Using a loosely coupled architecture Global Positioning System (GPS) and inertial navigation system (INS) are blended to provide estimates of the vehicle state. The system is implemented on the experimental vehicle to map various terrains. The terrain maps are characterized using RMSE, PSD, root mean squared slope (RMSS), and amplitude to wavelength ratio. Additionally, a feature extraction algorithm based on the wavelet transform is introduced. The response of the experimental vehicle on the terrains is analyzed using the RMS vertical acceleration, RMS roll rate, RMS pitch rate, and RMS suspension deflections. The 7-DOF suspension model of the experimental vehicle is then used to compare the simulated vehicle response to the experimental vehicle response. The model is then used to evaluate the effectiveness of the W-M function for generating artificial terrains. The response of the simulated vehicle on the experimental and generated terrains is then compared. It is determined that an artificial surface can be generated which will result in a similar vehicle response as the experimentally measured surface. The method does however have difficulty capturing the nuances of experimentally measured terrains. Additionally it is shown that the roughness of the terrain can be characterized by analyzing the surface with the RMSE, PSD, RMSS, or wavelength to amplitude ratio. The roughness can also be characterized by the RMS vertical acceleration, RMS roll rate, RMS pitch rate, or RMS suspension deflection. These methods are compared and their strengths and weaknesses are highlighted.