A Review on AI-Driven Approaches for Autonomous Vehicles: Progress and Challenges

Abstract

The rapid advancement of artificial intelligence has significantly propelled the development of autonomous vehicles, transforming both technological frameworks and practical applications. This paper systematically examines AI-driven approaches in autonomous vehicle systems, focusing on recent breakthroughs and persistent challenges. In perception systems, multi-sensor fusion and few-shot learning techniques have markedly enhanced object detection accuracy, while hierarchical reinforcement learning and socially compliant models have improved decision-making capabilities. Innovations in control systems, particularly the integration of model predictive control with neural-symbolic methods, demonstrate promising results in real-world scenarios. However, critical challenges remain, including performance degradation in extreme weather conditions, unresolved ethical and regulatory dilemmas regarding liability, and public skepticism toward human-machine interaction. The analysis highlights the necessity for explainable AI frameworks and real-time causal reasoning to address these issues. Future research directions emphasize the importance of cross-domain collaboration involving vehicle-road-cloud systems to achieve robust and trustworthy autonomous driving solutions. This review provides a comprehensive perspective on the current state of AI in autonomous vehicles, offering insights for researchers and practitioners navigating this evolving field.