IVLMap Solves Robot Navigation By Mapping Individual Objects

In this section, extensive experiments are presented to assess the suggested Instance-aware Visual Language Map (IVLMap) in both simulated and real-world settings. Leading baselines like VLMap, CoW, and CLIP Map were compared against IVLMap using the Habitat simulator and the Matterport3D dataset. Results show that IVLMap performs better in instance-aware and multi-object navigation tasks, allowing for zero-shot goal navigation and accurate localization.