This thesis is about design, implementing and testing a domain-independent, single-agent, deterministic and full observability suppose forward planner . The A.I. (Artificial Intelligence) is engaged with the understanding and creating of intelligent entities, within the planning discusses the modelling and the solutions of compound tasks. Through the planning there are applicable actions; if these actions are performed in a given order, the desired goal is achievable. With planning how to accomplish the source demanding tasks, the goals are achieved in a time and expense efficient way.
The language of input problem descriptions is the subset of the PDDL (Planning Domain Definition Language) 3.1. The PDDL is a (relatively international) standard for describing single-agent classical planning problems. The planner is hybrid and combines UCT (Upper Confidence Bounds for Trees) and A* search, in order to solve the problem, where the solution is a plan which is a series of actions.
The thesis displays the background theory of planning. After an overview of the PDDL language, the subset is chosen, which is supported by the implemented planner. Subsequently the description of the system’s plan is reviewed; the system has three main components: parser, builder and plan search engine. Their functions according to their sequence: syntactic analysis, semantic analysis along with creation of data structures, and the creation of the solution plan.
The system has been implemented in Java. The study summarizes the implementation-specific decisions, in addition it contains a brief guide how to use the application on Windows with JRE (Java SE Runtime Environment) 8.
The thesis includes a comparison of the implemented system and the winner planners of IPC (International Planning Competition) through simple problems and problems of IPC 2014. Afterwards the summary of the work, the thesis ends with displaying the possibilities of further development.