Path Planning with Large Language Models
- Traditional robot navigation has primarily relied on occupancy grid maps and laser-based sensing, as exemplified by the widely-used move_base ROS package.
- Unlike robots, humans navigate using not only spatial maps and physical distances but also incorporate external information such as elevator maintenance notifications from emails or experiential knowledge like the need for special access through certain doors.
- With the development of Large Language Models (LLMs), which can comprehend and process textual information, there is now an opportunity to infuse robot navigation systems with human-like understanding.
- In this project, we propose using osmAG (Area Graph in OpensStreetMap textual format), an innovative semantic topological hierarchical map representation to bridge the gap between move_base capabilities and contextual interpretation ability of LLMs.
- Our approach facilitates a more intelligent approach to robot navigation that leverages a broader range of informational inputs and yet still remain the robust capabilities of traditional robotic navigation.
Demonstration of path planning using Large Language Models