L2R-Nav: A Large Language Model-Enhanced Framework for Robotic Navigation

Robot navigation in dynamic, unfamiliar environments poses a significant challenge, as it traditionally relies on static maps, which are inadequate for the ever-changing scenarios encountered in daily life. This paper introduces L2R-Nav, an innovative, end-to-end intelligent robot navigation framework. It harnesses large language model technology combined with reinforcement learning to facilitate navigation tasks based on user instructions. L2R-Nav integrates the sophisticated cognitive abilities of large language models with the training of a local navigation model, employing a novel probabilistic graph approach. This integration is aimed at pioneering new methodologies in robot interaction and navigation. The robustness and effectiveness of the L2R-Nav framework are demonstrated through extensive empirical evaluations in a variety of environments, underscoring its potential as a significant advancement in the field of robotic navigation.