Research Project: | MobiNav3D - High-Precision Navigation for Mobile Robots through Photorealistic 3D Vision |  | |
| Research Area: | Visual Perception, Synthetic Data, Artificial Intelligence | ||
| Funded by: | IFB Hamburg | ||
| In collaboration with: |
| ||
| Project start: | January 2026 | ||
| Project end: | December 2027 |
MobiNav3D – High-Precision Navigation for Mobile Robots Through Photorealistic 3D Vision
The MobiNav3D project aims to develop a real-time, edge-optimized visual SLAM system for mobile robots. By using photorealistic 3D vision based on Gaussian splatting, manual analysis and inspection tasks will be automated.
A key innovation is the integration of previously isolated areas such as navigation, global referencing, and perception analysis into a single software solution. This integration significantly reduces latency and enables the creation of precise 3D environment models directly on embedded edge devices. The focus is on three main use cases: urban infrastructure mapping, interactive industrial inspection, and virtual commissioning.
IFPT contributes its extensive expertise in automation and AI-enabled visual inspection to this project. Within the project, the institute is primarily responsible for developing the perception algorithms. These algorithms autonomously detect relevant damage or anomalies based on the captured image data. To effectively train these AI models, the IFPT relies on an innovative combination of large fundamental models and synthetically generated data. Furthermore, the database will be expanded with AI-supported, but human-annotated, data. This hybrid approach of artificial and a small amount of real annotated data is used to further train the classifiers for segmentation and object recognition. This enables the algorithms to be enabled in a time- and cost-efficient manner, which is particularly advantageous in industrial environments.
 |
Contact person at the institute: M.Sc. Keno Moenck