Research on Cognitive Vision Architectures in Linkoping

by Per-Erik Forssen

This talk will describe recent work in the COSPAL project that I've been involved with.

A practically useful cognitive vision system has to be able do discover new knowledge. In practice the system cannot start from scratch, and learn from first principles, since this would take too long. In practice, thus there is always a mix between discovered and built-in, or innate knowledge that the system uses.

The COSPAL project aims at investigating different vision based cognitive architectures which relax the amounts of built-in knowledge that is needed. The success of a system design is evaluated on a shape-sorter puzzle. We are currently using a simulated 2D puzzle, and are moving to using an industrial robot and a 3D puzzle box.

In the talk I will give an overview of the project, and show some recent results.

Back to the LCI Forum page