Application of Transformer Models for Autonomous Off-Road Vehicle Control: Challenges and Insights

TY - CONF

T1 - Application of Transformer Models for Autonomous Off-Road Vehicle Control: Challenges and Insights

AU - Azhar, M A Hannan Bin

AU - Mészáros, Zoltán

AU - Islam, Tasmina

N1 - Publisher Copyright: © 2024 IEEE. M. A. H. B. Azhar, Z. Mészáros and T. Islam, "Application of Transformer Models for Autonomous Off-Road Vehicle Control: Challenges and Insights," 2024 International Conference on Intelligent Cybernetics Technology & Applications (ICICyTA), Bali, Indonesia, 2024, pp. 537-542, doi: 10.1109/ICICYTA64807.2024.10913212.

PY - 2025/3/11

Y1 - 2025/3/11

N2 - This paper addresses the critical challenge of advancing autonomous vehicle control in off-road environments, where traditional driver assistance technologies often prove inadequate. While AI-powered systems in modern vehicles have become highly effective at navigating structured urban landscapes, adapting these technologies for rural and off-road settings remains a complex and necessary undertaking due to varied and unpredictable obstacles. Off-road scenarios present unique challenges, such as dense vegetation, rugged terrain, uneven surfaces, and water bodies, which demand robust detection and classification capabilities beyond those found in urban areas. This study explores the application of state-of-the-art machine learning models, particularly transformer-based architectures, to enhance feature recognition and classification in rural contexts. We evaluate several advanced models, including hybrid architectures that combine convolutional neural networks (CNNs) with transformers, to determine their effectiveness in identifying complex off-road features. Findings reveal that, although current data limitations restrict the development of fully autonomous systems for off-road navigation, meaningful progress can still be achieved to improve driver assistance functionalities. This paper emphasises the urgent need for broader, more diverse datasets to ensure model robustness and generalizability for autonomous navigation in unstructured, unpredictable environments. Ultimately, this work highlights a promising path toward safer, more effective driver assistance technologies tailored specifically for challenging off-road applications and scenarios.

AB - This paper addresses the critical challenge of advancing autonomous vehicle control in off-road environments, where traditional driver assistance technologies often prove inadequate. While AI-powered systems in modern vehicles have become highly effective at navigating structured urban landscapes, adapting these technologies for rural and off-road settings remains a complex and necessary undertaking due to varied and unpredictable obstacles. Off-road scenarios present unique challenges, such as dense vegetation, rugged terrain, uneven surfaces, and water bodies, which demand robust detection and classification capabilities beyond those found in urban areas. This study explores the application of state-of-the-art machine learning models, particularly transformer-based architectures, to enhance feature recognition and classification in rural contexts. We evaluate several advanced models, including hybrid architectures that combine convolutional neural networks (CNNs) with transformers, to determine their effectiveness in identifying complex off-road features. Findings reveal that, although current data limitations restrict the development of fully autonomous systems for off-road navigation, meaningful progress can still be achieved to improve driver assistance functionalities. This paper emphasises the urgent need for broader, more diverse datasets to ensure model robustness and generalizability for autonomous navigation in unstructured, unpredictable environments. Ultimately, this work highlights a promising path toward safer, more effective driver assistance technologies tailored specifically for challenging off-road applications and scenarios.

UR - http://www.scopus.com/inward/record.url?scp=105001659642&partnerID=8YFLogxK

U2 - 10.1109/ICICYTA64807.2024.10913212

DO - 10.1109/ICICYTA64807.2024.10913212

M3 - Paper

SP - 537

EP - 542

T2 - 2024 International Conference on Intelligent Cybernetics Technology & Applications (ICICyTA)

Y2 - 17 December 2024 through 19 December 2024

ER -