FASIONAD combines end-to-end planners with vision-language models in a dual-system approach inspired by human cognition to address safety challenges in autonomous driving.

• Utilizes a "fast" traditional planner for routine driving scenarios
• Engages a "slow" VLM-based reasoning module for complex situations
• Implements information bottleneck techniques to improve computational efficiency
• Employs adaptive feedback mechanisms for continuous improvement

This research represents a significant advancement for autonomous vehicle engineering by addressing the critical balance between efficiency and safety, particularly in handling edge cases and complex driving scenarios that traditional systems struggle with.

FASIONAD++: Integrating High-Level Instruction and Information Bottleneck in FAt-Slow fusION Systems for Enhanced Safety in Autonomous Driving with Adaptive Feedback