End-to-End Autonomous Driving: Challenges and Frontiers

Area of Science:

• Robotics
• Artificial Intelligence
• Computer Science

Background:

• The autonomous driving field is increasingly adopting end-to-end frameworks.
• These systems process raw sensor data for direct motion planning, unlike modular approaches.
• Joint optimization of perception and planning offers advantages over task-specific modules.

Purpose of the Study:

• To provide a comprehensive survey of end-to-end autonomous driving research.
• To analyze the motivation, methodology, challenges, and future directions in the field.
• To cover over 270 relevant academic papers.

Main Methods:

• Systematic literature review of end-to-end autonomous driving papers.
• Analysis of methodologies, including perception, planning, and sensor fusion.
• Discussion of challenges such as multi-modality, interpretability, and robustness.

Main Results:

• Identified key trends and advancements in end-to-end autonomous driving.
• Detailed analysis of over 270 papers, categorizing their approaches and findings.
• Highlighted critical challenges and potential solutions within the surveyed literature.

Conclusions:

• End-to-end driving systems show significant promise, benefiting from joint optimization.
• Addressing challenges like robustness and interpretability is crucial for future development.
• Emerging techniques like foundation models offer new avenues for end-to-end driving frameworks.