From Coarse Sampling to Insight: An Intelligent Testing Technique for Autonomous Haulage Systems

Abstract

The mining industry is undergoing rapid transformation through the adoption of autonomous haulage systems (AHS), which promise improved safety, efficiency, and productivity. However, the testing of these systems remains a critical yet underdeveloped area. Based on a thematic analysis of interviews with domain expert engineers, this thesis investigates the challenges of testing AHS and examines the use of intelligent testing in virtual simulations to support their development. Coarse gridsearch testing, combined with adaptive sampling strategies and tree-based models, enables feature importance analysis, allowing us to isolate scenario parameters Autonomous Haulage System,Autonomous Haulage System,Autonomous Haulage System,Autonomous Haulage System,Autonomous Haulage System,Autonomous Haulage System,Autonomous Haulage System,Autonomous Haulage System, Autonomous Haulage System, that are largely independent of system performance. This dimensionality reduction also facilitates the construction of easy-to-understand, shallower decision trees and simplified random forests that can identify parameter regions likely to violate system specifications. Additionally, we identify input pairs with minimal differences that result in maximal output discrepancies, helping developers and testers alike to quickly localize error-inducing components. Our initial results demonstrate the value of intelligent testing as a proactive tool for uncovering edge cases, going beyond conventional scenario replay and widely used expert-driven methods. This thesis contributes to the development of a more robust and systematic testing framework for AHS, aiming to improve their reliability and operational safety.