Abstract: This paper presents a novel framework for memory-based navigation for terrestrial robots, utilizing a customized multimodal large language model (MLLM) to interpret visual inputs and ...