Neural network enhanced off-road skid-steer vehicle modeling with an application to path planning
Yurkanin, Justin
This item's files can only be accessed by the System Administrators group.
Permalink
https://hdl.handle.net/2142/115870
Description
Title
Neural network enhanced off-road skid-steer vehicle modeling with an application to path planning
Author(s)
Yurkanin, Justin
Issue Date
2022-06-28
Director of Research (if dissertation) or Advisor (if thesis)
Norris, William R
Ramos, Joao
Department of Study
Mechanical Sci & Engineering
Discipline
Mechanical Engineering
Degree Granting Institution
University of Illinois at Urbana-Champaign
Degree Name
M.S.
Degree Level
Thesis
Keyword(s)
Autonomous
Off-road
neural ODE
dynamics
Abstract
This thesis discusses the development and validation of 2D and 3D dynamic skid-steer vehicle models for the purpose of enabling or facilitating the future development of control, path planning, and localization algorithms off-road. The ideal dynamic model is fast, accurate, general, adaptive, and 3D. The work presented in this thesis tries to approach this ideal with different models. First, a very simple linear 2D vehicle model was trained from data as a benchmark. Following this, a more complex 2D neural network model was developed. Next, a 3D floating base dynamic vehicle model was created. This was integrated with the Bekker tire-soil model which was approximated with a neural network and used to realistically simulate terrain. Auto differentiation was leveraged to create a fully differentiable 3D simulation which enabled optimization techniques. Gradient descent was used to select soil parameters that maximized 3D model accuracy and achieved offline model adaption to unknown soil types. Finally, experiments were performed to train the 3D vehicle model as a physics based neural ODE. All vehicle models were trained and evaluated with an external data set. To demonstrate the usefulness of the 3D model, a Rapid Random Trees (RRT) algorithm was implemented in simulation to search for valid paths across a 3D terrain.
Use this login method if you
don't
have an
@illinois.edu
email address.
(Oops, I do have one)
IDEALS migrated to a new platform on June 23, 2022. If you created
your account prior to this date, you will have to reset your password
using the forgot-password link below.