USING SIMULATION IN THE DESIGN, TESTING, AND OPTIMIZATION OF AUTONOMY STACKS FOR AUTONOMOUS GROUND VEHICLES

Abstract

The integration of autonomy in an increasing number of engineering systems, e.g., rovers, all terrain ground vehicles, quadcopters, calls for effective tools that can assist one in developing robust and efficient autonomy stacks (A-stacks). This thesis explores the use of simulation technologies to enhance the design, validation, and optimization of A-stacks for autonomous ground vehicles. Leveraging the Chrono simulation platform, which can model the vehicle dynamics and sensor operation, this work underlines the critical role that simulation can play in autonomy applications. The thesis highlights the design of A-stacks through simulation, utilizing artificial intelli- gence (AI) techniques to generate and refine control policies. The innovative use of Chrono’s terramechanics simulations is a strength of this work, providing unique insights into the vehicle-environment interaction. By leveraging simulations in a diverse collection of scenarios to increase robustness, the development process reduces the reliance on extensive physical testing, thus optimizing resource utilization and accelerating the development cycle. The thesis also investigates the process of validation of simulation-designed A-stacks. By employing techniques such as a priori predictions and a posterior measurements, this research assesses the sim-to-real transferability and the predictability of A-stack performance. The work suggests that simulation in not only useful to design the autonomy algorithms, but can also be used to predict the performance of a particular A-Stack before being deployed in the real world. This is accomplished through a so-called “random testing” regimen.