Abstract
Simultaneous Localization and Mapping (SLAM) is a cornerstone of robotics and AR/VR applications. While SLAM simulations are essential for testing new approaches, resource-constrained devices such as VR head-mounted displays (HMDs) pose challenges due to high computational demands and limited access to sensor data. This work introduces a sparse SLAM framework that leverages mesh geometry projections as features, improving efficiency and bypassing the need for direct sensor input. We demonstrate its effectiveness in simulation for VR applications and through numerical evaluations, advancing SLAM research for constrained environments.
About the speaker
Carlos Pinheiro is a PhD candidate at the University of Konstanz, researching real-time SLAM and 3D reconstruction for XR, robotics, and multi-agent systems. His background in CGI, video game development, and computer vision bridges graphics and vision with classical and modern AI—expertise he now applies to enabling efficient spatial computing on resource-constrained devices. Before re-entering academia, he worked at Softkinetic (now Sony DepthSensing Solutions), NXP Semiconductors, MBJ Solutions, Hensoldt Sensors, and Lake Fusion Technologies—where he developed systems for embedded validation, industrial AI, military VR, multi-object tracking on different domains including LiDAR, camera, and radar, and mixed-reality ADAS simulation.