Masters Thesis Presentation - Alexandru Mihai Chirita

Title: Multi-modal odometry and semantic mapping

Abstract:

Simultaneous Localization and Mapping (SLAM) is a foundational method in robotics that enables a robot to build a map of its environment while localizing itself within it. Recently, there has been a growing interest in indoor scene understanding and 3D mapping, which can enable autonomous real-world task completion and provide a common interface for more intelligent human-robot interactions. While current research heavily focuses on geometric 3D mapping, pure geometry is insufficient for constructing a truly comprehensive and actionable model of the surrounding environment.

To bridge this gap, we propose a loosely coupled, multi-modal odometry and semantic mapping system. Our approach is based on GLIM, a 3D range–inertial localization and mapping framework, and incorporates GPU-accelerated scan matching factors together with the instance segmentation head of the YOLO model. By fusing these modalities, we achieve a highly efficient system capable of segmenting 3D point clouds in real time.

Supervisors: Ayushi Rastogi, Kailai Lee