A fundamental prerequisite for robotic grasping of objects and, more generally, sensible interaction with objects in the environment is knowledge of the objects’ identity and pose, the latter relative to some object-related reference frame. When dealing with objects from a known finite set, these perceptual faculties are generally called object recognition and pose estimation. Many robotic applications in assembly, logistics, and service domains depend on that ability.
We offer a research project for a master student to investigate classification and regression techniques for the purpose of object recognition and pose estimation. The objects and scenes to be analyzed will be given as RGBD data from public repositories. The methods to be investigated will be based on new variants of Random Forests.
Start date: as soon as possible