From Language to Navigation Goals: A Vision-Language Approach for Semantic Navigation of Mobile Robots Using RGB-D Perception

The paper presents a modular ROS 2 framework that converts natural-language navigation requests, such as “go to the mail box,” into executable goals for mobile robots. The system identifies the referenced object, estimates its position from RGB-D perception, and sends a navigation goal to the ROS 2 Nav2 stack. Evaluation covers simulation and real-world experiments with a TurtleBot3 Waffle and a Unitree Go2 equipped with a RealSense camera. The authors report successful handling of direct and contextual requests, along with natural-language feedback. Source code is planned for release after acceptance.