Decreasing the Reality Gap of a Vehicle Simulation Digital Sibling Using the Addition of a Road Slope Study
About this thesis
This thesis focuses on reducing the reality gap in vehicle simulation platforms by integrating a comprehensive study of road slopes. Simulation platforms are critical tools for developing and testing driver-assistance systems (ADAS) and autonomous vehicles, offering a safe, cost-effective alternative to real-world tests. However, achieving high levels of realism, especially in road and vehicle dynamics, remains a significant challenge.
By incorporating detailed road slope data, this work aims to bridge the gap between simulated and real-world driving conditions. The enhanced realism enables more accurate testing of ADAS and autonomous driving technologies. This thesis delves into how custom soft-body physics engines and adaptable road models contribute to more reliable and immersive simulations, making it ideal for both academic and industry applications.
Introduction
As the complexity of modern driver training systems, ADAS, and autonomous driving technologies continues to grow, the demand for hyper-realistic simulation platforms becomes more pronounced. These platforms are indispensable for safely testing complex driving scenarios without exposing vehicles or drivers to real-world risks. However, one of the significant hurdles is accurately modeling real-world conditions such as road geometry, traffic patterns, and environmental factors like slopes.
This research leverages platforms like BeamNG.tech and Hexagon’s Virtual Test Drive (VTD) to improve the accuracy of road simulations. A major innovation of this thesis is the integration of OpenStreetMap (OSM) road data with a specific focus on road slopes, a critical factor in vehicle dynamics. This addition allows for a more realistic replication of how vehicles interact with varying terrain, which is essential for ADAS testing and the development of autonomous agents.
By focusing on the customization of simulation inputs, this research enhances the ability of simulation platforms to more closely mirror real-world driving environments. The study compares the performance of standard road data imports against customized imports, ultimately demonstrating how improved realism can lead to better development and testing outcomes.