Student Theses

Entwicklung einer hochautomatisierten Methode zur Erstellung gekoppelter fotorealistischer Simulationsumgebungen im Kontext mobiler Robotik

Mobile robots increasingly rely on vision-based algorithms, yet their development in simulation is hindered by the visual discrepancy between simulated and real camera images. 3D Gaussian Splatting enables photorealistic rendering from real image data in real time, offering a promising approach to overcoming this so-called Sim-to-Real Gap. The goal of this thesis was to enable a workflow in which a mobile robot drives through a scene once and the recorded sensor data is automatically turned into a photorealistic Gazebo simulation of the same robot in the same environment. To this end, a modular reconstruction pipeline was developed that creates a dual scene representation consisting of a collision mesh and a 3D Gaussian Splatting model as well as the corresponding Gazebo simulation world based on a SLAM reconstruction. A rendering backend integrated into Gazebo couples the physics simulation with the photorealistic rendering through an asynchronous architecture, so that the camera pose for rendering is obtained directly from the physics simulation. Evaluation on five self-recorded datasets showed that the pipeline generates all simulation files automatically and that the integrated rendering easily achieves real-time capability. The system thus enables the path from a single data recording to photorealistic robotics simulation in Gazebo.

Anomalieerkennung in Kleegras mittels maschinellem Lernen

Fusion of Distance and Hyperspectral Sensor Data for the Generation of Synthetic Plant Training Models

Agriculture today faces many economic and environmental challenges. Among other things, the increasing digitalisation of agriculture is seen as a basis for the further development of agriculture to meet these challenges. Artificial intelligence and machine learning can be seen as important core technologies. However, in order to train their algorithms, large amounts of training data are required, the data of which are usually difficult to collect under real conditions. Therefore, this work develops an open-source visual inspection system to capture and fuse geometric and spectral information under laboratory conditions is developed to generate synthetic plant training models. For this purpose, two laser-lineprofilers and a Pushbroom hyperspectral camera are combined in a benchtop experimental setup. The generated datasets of the recorded plant test objects are then algorithmically processed, merged into hyperspectral point clouds and exported as textured meshes for further use. Finally, the results of the plant samples and the developed visual inspection system are evaluated, discussed and validated. The evaluations show that the developed visual inspection system is more powerful than those of comparable scientific work.