This research presents a dual-core framework that leverages large language models to overcome limitations in traditional navigation systems when handling non-predefined scenarios.

• Uses ReAct-based prompt engineering to decompose complex navigation tasks into manageable sub-tasks
• Implements a larger LLM core for complex reasoning and task decomposition
• Enables autonomous invocation of external tools to adapt to unknown navigation conditions
• Enhances safety and reliability in navigation applications

For engineering teams, this framework represents a significant advancement in creating more flexible, intelligent navigation systems that can adapt to unpredictable real-world conditions rather than being limited to predefined scenarios.

Navigation-GPT: A Robust and Adaptive Framework Utilizing Large Language Models for Navigation Applications