The thesis belongs to the field of Artificial Intelligence, robotics and qualitative reasoning. The purpose of the work is to use a qualitative simulator for planning qualitative actions of a robot. Our modification of the known QSIM algorithm generates state space, which we search with the heuristic search algorithm A*. Implementations of all algorithms are written in the programming language Prolog. Some machine learning algorithms induce qualitative models using QDE constraints that are not defined in the original QSIM algorithm. One of these QDE constraints is the monotonicity in multiple variables. This QDE constraint was implemented and tested on an artificial domain. Generated robot plans have been tested on an object pushing simulator, which is based on the Box2D engine. For this purpose, an algorithm for plan execution was developed. This plan execution algorithm communicates through an interface, which was also developed as part of the thesis. The interface is responsible for a conversion of numerical data into qualitative states. The interface also implements execution of qualitative actions on the simulator. Plans developed by the proposed algorithm have been tested in two object pushing domains: the case of pushing a vertical cylinder and the case of pushing a block. For this purpose, there was a hand-built qualitative model for each domain. The thesis is concluded with an examination of achieved objectives, a review of potential challenges in the implementation of algorithms and a review of ideas for further research.
|
