Incomplete observations
Shape completion, uncertainty estimation and persistent mapping for scenes a robot cannot fully observe.
Active perception · scene reasoning · manipulation
Prior-guided active perception for scene understanding and manipulation.
I am a robotics researcher and PhD candidate at the University of Bonn. I develop systems that combine learned priors, uncertainty, scene graphs and geometric planning to decide what a robot should observe and do next.
Before my PhD, I built robot perception, manipulation and hard real-time control systems at NEURA Robotics and DFKI. I also teach robotics, supervise student research and present robots in public demonstrations.
Robot demonstration
HortiBot. A 2× preview of active perception, online plant mapping and manipulation on the three-arm horticultural robot. Watch the full paper video.
Research themes
Reasoning and active perception
Scene graphs, object search and view planning that connect semantic priors to physical actions.
Perception–action integration
Real mobile manipulators, multi-arm systems and shared autonomy validated beyond simulation.
Selected updates
SG-AMP preprint
Scene-graph-guided active perception with uncertainty-aware depth completion and semantics-aware motion planning.
Read the preprintTwo thesis awards
Two supervised MSc theses received the Bonner Informatik Group Best Thesis Award and the Grace Hopper Thesis Award.
HRL newsIROS workshop award
Best Poster Award for open-vocabulary semantic search in scene graphs—a distinct precursor to the later O-STaR work.
Award announcement