Feature based Object Relational Navigation

The aim of this DFG-funded research project is to develop a system for object-relational navigation that does not follow global trajectories in a previously created metric map, but rather navigates relative to perceived objects ("at the third intersection left, past the tree the second house right"). Furthermore, the inclusion of object features (prominent edges, textures, etc.) in the navigation should increase robustness compared to current approaches that only use abstracted box models with artificial reference points. By linking navigation to perceived objects, the aim is on the one hand to guarantee a quick reaction to unexpected behaviour, such as the sudden departure of the vehicle on the side lane, and on the other hand to reduce the complexity of plan generation and ongoing plan changes, especially in dynamic scenes, through plans such as: "Follow road, overtake vehicle X,...".

Projektseite der DFG