Fuzzy-genetic hybrid models for large-horizon deterministic planning

Identifier:  TFM:2341

Handle: https://hdl.handle.net/20.500.11797/TFM2341

Authors:  Safronov, Mark

Abstract:
This thesis addresses deterministic planning problems with long horizons, motivated by the life simulation game Princess Maker 2. The formulation involves an agent with dozens of attributes, several dozen actions, and thousands of steps, leading to combinatorial explosion. Classical approaches such as automatic planning and reinforcement learning fail to scale. We propose a fuzzy-genetic hybrid method: fuzzy logic encodes domain knowledge as rules over latent parameters (“inclinations”), while genetic algorithms optimize them. Implemented in C++ with fuzzylite and pagmo, the solver successfully produces valid action sequences, demonstrating tractability where classical methods are impractical.