Simultaneous Localisation and Mapping (SLAM) is a fundamental problem in robotics, involving the estimation of both the pose of a mobile robot and the map of its surroundings. This representations can then be used by the robot to make and execute decisions. Recent advancements address the SLAM problem by employing a graph-based representation that integrates metric (occupancy) and semantic (classes) estimates, along with hierarchical relationships. This enhanced representation, known as a 3D Dynamic Scene Graph (DSG), encompasses actionable information that aids in robot planning and decision-making, but also in supporting architecture-related tasks such as renovation or remote building visualisation. For instance, Hydra [1] leverages 3D geometry to cluster free space into a graph of places, topology to group these places into rooms, and geometric deep learning to classify the types of rooms the robot navigates through.

During this project, the student will deploy and compare three different DSG methods from the state of the art and evaluate their performance in a simulated environment. Finally, the best performing method will be used to process real world data and create a DSG of the Human Centred Computing and Robotics (HCCR) laboratory. Implementation will be done using the Robotics Operating System (ROS2).

[1] Hughes N, Chang Y, Hu S, et al. Foundations of spatial perception for robotics: Hierarchical representations and real-time systems. The International Journal of Robotics Research. 2024;43(10):1457-1505.