70 likes | 169 Views
Thuggabot. THe University of Georgia Genetic Algorithm BOT. Thuggabot. Half-Life Game World Thuggabot Concepts Combat Strategy Genetic Algorithm Learning Test Results Demo. Half-Life Game World. First-Person Shooter (3D Environment) Objective: Maximize kills, Minimize Deaths
E N D
Thuggabot THe University of Georgia Genetic Algorithm BOT
Thuggabot • Half-Life Game World • Thuggabot Concepts • Combat Strategy • Genetic Algorithm Learning • Test Results • Demo
Half-Life Game World • First-Person Shooter (3D Environment) • Objective: Maximize kills, Minimize Deaths • Upon dying, players re-spawn with minimal equipment. • Throughout the game, players gather items to help them accomplish goals.
Thuggabot Concepts • AI Combat Agent • Acts to simulate human player • Goal Oriented • Utilizes Genetic Algorithm • Based on the HPB Bot Framework by Botman
Combat Strategy • Each bot has preferences regarding possible actions • Bots choose goals based on preferences • Bots which make good choices are more effective in combat • Bots adapt to their environment through evolution.
Genetic Algorithms • Representation • Array of weights that correspond to actions and weapon preferences • Proportional Fitness Tournament Selection • Uniform Crossover • Random Index Mutation
Test Results • Roughly monotone increasing performance • Some goals clearly become favored over others • Some preferences fluctuate due to dynamic nature of the environment. • Tested against TheFatal’s “Jumbot,” Thuggabot achieved long-term domination