This investigation explores the incorporation of sophisticated large language models (LLMs) into HighwayEnv for autonomous driving (AD), aiming to replicate intricate human-like driving behaviors, overcoming the limitations inherent in traditional AD systems. Renowned for their sophisticated learning capabilities and natural language processing, LLMs provide innovative solutions to the rigidity and predictability challenges faced by conventional AD frameworks, particularly in adapting to unexpected driving conditions. This study highlights the LLMs’ proficiency in interpreting complex driving environments through analysis of sensor data, traffic signals, and other critical road indicators, mirroring human perception and decision-making processes. By leveraging past driving experiences, LLMs enhance their predictive accuracy and adaptability, crucial for dynamic decision-making in AD systems. We construct an AD framework within the HighwayEnv, showcasing LLMs’ adeptness at navigating intricate scenarios, characterized by variable weather conditions, road geometries, and traffic densities. Extensive testing underlines the LLMs’ competence in logical deduction and managing rare driving events, demonstrating their capacity to emulate sophisticated human driving behaviors. The findings of this study underscore the significant potential of LLMs to enhance the safety and dependability of autonomous vehicles, especially in navigating the complexities of driving situations. This paper posits that LLMs represent a significant leap forward in the quest for fully autonomous, human-like driving within the structured yet diverse scenarios presented by HighwayEnv, marking a pivotal advancement in autonomous vehicle technologies.
Advancing Autonomous Driving with Large Language Models: Integration and Impact
2024-05-09
501496 byte
Conference paper
Electronic Resource
English