DSM (Diverse Semantic Map) enhances robotic vision by creating richer 3D scene understanding for more accurate object identification and interaction.

• Extracts diverse semantic information from visual scenes, including implicit attributes often missed by current systems
• Improves 3D Visual Grounding capabilities, helping robots better locate and interact with specific objects
• Builds upon multimodal large language models (VLMs) to create more comprehensive scene representation
• Addresses limitations in existing approaches that rely primarily on geometric and visual data

This research significantly advances robotics engineering by enabling more natural human-robot interaction through improved object recognition and contextual understanding in complex environments.

DSM: Building A Diverse Semantic Map for 3D Visual Grounding