MORE | Spring 2025

Real-time 6D Pose Tracking of Novel Objects Using 2D Gaussian Splatting

Real-time 6D pose tracking of novel objects using a single camera is challenging due to slow reconstruction, occlusion sensitivity, and the need for extensive training. Existing methods take about 12 minutes for reconstruction and operate around 13 Hz for 6D pose tracking of the object. To overcome these challenges, a real-time tracking method using 2D Gaussian splatting, enabling fast object reconstruction and tracking via rendering loss optimization, is proposed. This approach eliminates the need for prior models and extensive datasets while maintaining accuracy. This method enhances robotic manipulation in dynamic environments, with applications in manufacturing, assistive and surgical robotics.