In order to be of use to humans, service robots must have a certain level of intelligence. They must be able to identify objects clearly and relate different spatial views to each other. This is exactly what researchers at the University of Bayreuth have developed a system for in the "SeLaVi" project.

Robots learn to see

"SeLaVi" stands for "Semantic and Local Computer Vision based on Color/Depth Cameras in Robotics". The project is led by Prof. Dr. Dominik Henrich at the Chair of Robotics and Embedded Systems. The aim is to enable robots to recognize what is happening in their environment on the basis of a few characteristic images. Cameras, which are attached to the arms of the robots like eyes, play a central role in this. The cameras generate images of the objects in their working environment, which are translated into geometric models. These models are characterized by the fact that they only depict a few representative surface areas of the objects. They are therefore called boundary representations (BReps) and require less memory and computing capacity than the point clouds or triangular meshes that have been commonly used for object recognition up to now. The new models are merged with additional color information contained in the camera images and stored in a database. By comparing new camera images with the database, the robots can recognize objects in their environment without errors. They are not disturbed by movements of neighboring objects.

This ability of the robots forms the basis for further learning steps, which concern the meaningful relationships between the static or moving objects in their working environment. Robots should learn to understand these "semantic relations" so that they can then use their arms to intervene in the respective scenarios in an appropriate manner. All of these capabilities, as they are currently being developed and optimized in Bayreuth, can be used in many areas - from autonomous service robots to cooperation between humans and robots.

According to documents from the University of Bayreuth / ag