Saf3D — Bachelor Thesis

Saf3D is my Bachelor thesis at BFH, supervised by Prof. Dr. Christoph Paulus. The goal is to reconstruct the physical environment of a robot on the Auto-Mate platform as a 3D digital twin, so that motion planning and safety can be based on a current, real-world scene.

The pipeline combines two approaches: COLMAP-based structure-from-motion for initial pose estimation and Brush for the actual Gaussian splatting. The final .ply splats are transferred via gRPC to the VRobot.

Stack: Python, COLMAP, Brush (Gaussian Splatting), gRPC.