Machine Learning for Real Time Navigation

February 9th 2002 Machine Learning Delphi

This project demonstrates the use of machine learning methods (genetic algorithms and nearest neighbors algorithm in particular) in real time navigation in unknown environment with given constraints. Apart from complete implementation with Delphi sources included (detailed explanation below) the project also features lots of interesting ideas in regard to attributes collection needed for such a navigation.

For the needs of the project I have developed the simulation environment (simulating the movement of an agent in a 2D map) with a nice GUI and a built-in editor with map management that could be of use even for other projects. More specific is a very flexible object oriented implementation of the genetic algorithms and a simple implementation of basic nearest neighbors algorithm. All the source code is nicely organized and laid out and as such suitable for reuse and extending.

The archive also contains some sample data which directly demonstrates the really impressive results that where achieved with these methods and a quite extensive technical report which is unfortunately written in Slovene as I had to submit it at university. The source code should be completely usable even without understanding this document but you'll of course miss some of the explanations.

Creative Commons License