We have built a system to make experiments with robots easy. The scenario is a camera mounted in the ceiling surveying the scene on which two robots move. They can communicate between themselves and with a central PC via a broadcast radiolink. The robots can be autonomously programmed or can be guided by processes on the PC. The camera proces analyzes the scene and transmits the robots' positions to the robots.

The success criterium for the class is that the students initiate design and implementation of   non-trivial behavior for a mobile robot. It is realistic that more time is needed for a  satisfactory result than available at this class. The students are encouraged to continue their projects after termination of the present class.

Two different architectures are advocated in this course. They both have a common module.  This module can be implemented in two ways. It is thus possible to initiate work on common grounds and differentiate to a specific project during the course.

The common first module of both proposed designs are to implement a ''vision'' module for  robot. The word vision is quoted, since I suggest 1-dimensional and not 2-dimensional vision. This simply for engineering feasability. . This module can be implemented either using the ceiling-mounted camera or it can be based on data from the on-board camera and the infrared sensors. This last approach is based upon the locomotion strategy of the house fly and invokes basic pattern recognition principles.

One design then follows the design of Tu and Terzopoulos, with the important exception that our system function in the physical real world, where Tu's system functions in the virtual world. This design incorporates the exiting ideas of Reynolds and builds upon Brook's subsumption architecture. This design follows the main stream in AI.

Another design will also be presented. It is based on information theory and string matching. It is connected to data compression and raises an interesting algoritmic problem. It is also more closely connected to biological vision.

---

Kommentarer til denne side bedes rettet til Peter Johansen