LetsPi introduces a novel dual-phase architecture that enhances transportation safety planning by combining physics principles with transportation-specific knowledge.

• Overcomes common LLM limitations including hallucinations and high latency in safety-critical contexts
• Integrates physics-informed social force dynamics into trajectory planning
• Provides transportation-specific safety knowledge beyond general collision avoidance
• Creates a more reliable framework for autonomous navigation in complex environments

This research significantly advances engineering safety in autonomous systems by embedding domain-specific knowledge directly into planning systems, potentially reducing accidents and improving reliability in real-world applications.

Planning Safety Trajectories with Dual-Phase, Physics-Informed, and Transportation Knowledge-Driven Large Language Models