Research Analyzer
← Back ICRA 2026

DualMap: Online Open-Vocabulary Semantic Mapping for Natural Language Navigation in Dynamic Changing Scenes

Jiajun Jiang, Yiming Zhu, Zirui Wu, Jie Song

PDF

AI summary

Key figure (auto-extracted from paper)
DualMap enables real-time, language-guided robot navigation in dynamic environments by combining a detailed local map with a simplified global map for efficient planning and adaptation.
Open-vocabulary mapping Dynamic navigation Semantic mapping Natural language navigation Dual-map representation Online robotics

Problem

Existing semantic mapping systems struggle to simultaneously support open-vocabulary queries, efficient online updates, and navigation in dynamically changing environments, limiting their real-world applicability.

Approach

The system builds a detailed 3D concrete map using a hybrid segmentation frontend, then abstracts it into a global map of static anchors for high-level planning, allowing online updates and re-navigation when objects move.

Key results

  • State-of-the-art performance in 3D open-vocabulary segmentation and online mapping
  • Elimination of costly 3D inter-object merging via lightweight intra-object status checks
  • Real-time language-guided navigation that adapts to dynamic object relocations
  • Validated across simulation benchmarks and real-world indoor/outdoor experiments

Why it matters

Enables practical, lifelong robotic navigation in human-centered spaces where objects frequently move and queries are open-ended.

Abstract

We introduce DualMap, an online open-vocabulary mapping system that enables robots to understand and navigate dynamically changing environments through natural language queries. Designed for efficient semantic mapping and adaptability to changing environments, DualMap meets the essential requirements for real-world robot navigation applications. Our proposed hybrid segmentation frontend and object-level status check eliminate the costly 3D object merging required by prior methods, enabling efficient online scene mapping. The dual-map representation combines a global abstract map for high-level candidate selection with a local concrete map for precise goal-reaching, effectively managing and updating dynamic changes in the environment. Through extensive experiments in both simulation and real-world scenarios, we demonstrate state- of-the-art performance in 3D open-vocabulary segmentation, efficient scene mapping, and online language-guided navigation. Project page: https://eku127.github.io/DualMap/

Index terms

Mapping Semantic Scene Understanding Vision-Based Navigation

Related papers