ManeuverGPT introduces a framework where LLM-based agents control autonomous vehicles to execute professional-level evasive maneuvers at the handling limits.

Key Innovations:

• Enables autonomous vehicles to perform high-dynamic stunt maneuvers for enhanced safety
• Utilizes large language models as controllers for complex driving scenarios
• Targets aggressive evasive maneuvers similar to those executed by professional stunt drivers
• Pushes the boundaries of vehicle handling capabilities in autonomous systems

Engineering Impact:

This research represents a significant advancement in autonomous vehicle safety systems by enabling dynamic responses to hazardous situations beyond traditional control methods, potentially saving lives in scenarios where standard evasive actions would be insufficient.

ManeuverGPT Agentic Control for Safe Autonomous Stunt Maneuvers