KakuroAI : διαδικαστική παραγωγή περιεχομένου με προσαρμοστική δυσκολία χρησιμοποιώντας ενισχυτική μάθηση
KakuroAI : procedural content generation with adaptive difficulty using reinforcement learning
View/
Keywords
Kakuro ; Procedural Content Generation (PCG) ; Proximal policy optimization ; Reinforcement learning ; Unity ML-agents ; Adaptive difficulty ; Puzzle generationAbstract
This thesis presents KakuroAI, a system that combines Procedural Content
Generation (PCG) with Reinforcement Learning (RL) to generate adaptive
difficulty Kakuro puzzles. Kakuro puzzles are logic-based number puzzles that
require players to fill a grid while following specific sum constraints. Despite their
popularity, generating Kakuro puzzles with a targeted difficulty level remains a
challenge. Our system addresses this issue by employing an RL agent trained to
adjust puzzle generation parameters based on player performance metrics.
The thesis describes the system’s architecture, the puzzle generation algorithm,
and the RL model used to adapt the difficulty. The system was implemented
using the Unity ML-Agents framework and the Proximal Policy Optimization
(PPO) algorithm. It monitors various player performance metrics, including
solving time, number of mistakes, and usage of hints, to model player ability and
dynamically adjust the puzzle difficulty.
We evaluated the system through experiments with real players and simulated
gameplay sessions. The results show that KakuroAI can effectively generate
Kakuro puzzles that adapt to the player's skill level, maintaining a balance
between challenge and solvability. Furthermore, the system demonstrates
improved performance compared to traditional puzzle generation approaches,
producing puzzles with more consistent difficulty levels and better alignment with
individual player capabilities.
This work contributes to the field of Procedural Content Generation and adaptive
games, introducing an innovative approach for generating puzzles with
personalized difficulty. The techniques developed in this thesis can be extended
to other types of puzzles and games, enhancing player experience and
accessibility.
Postgraduate Studies Programme
Προηγμένα Συστήματα Πληροφορικής - Ανάπτυξη Λογισμικού και Τεχνητής ΝοημοσύνηςDepartment
Σχολή Τεχνολογιών Πληροφορικής και Επικοινωνιών. Τμήμα ΠληροφορικήςNumber of pages
63Language
GreekCollections
Except where otherwise noted, this item's license is described as
Αναφορά Δημιουργού - Μη Εμπορική Χρήση - Παρόμοια Διανομή 3.0 Ελλάδα
Αναφορά Δημιουργού - Μη Εμπορική Χρήση - Παρόμοια Διανομή 3.0 Ελλάδα