LetsPi introduces a dual-phase architecture that combines physics models with transportation knowledge to overcome hallucination and safety challenges in autonomous trajectory planning.

• Leverages physics-informed social force dynamics to create realistic trajectories
• Uses a two-phase approach: knowledge-driven planning followed by physics-based refinement
• Addresses key limitations of standard LLMs: hallucinations, uncertainty, and inference latency
• Incorporates domain-specific safety knowledge for transportation scenarios

This research represents a critical advancement for engineering safer autonomous systems by integrating physical constraints and specialized knowledge into LLM reasoning—potentially reducing safety incidents and improving real-world deployment viability.

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