IEEE AESS - St. Mary's - Survivability and Combat Tactics in StarCraft II : vTools Events

IEEE.org | IEEE Xplore Digital Library | IEEE Standards | IEEE Spectrum | More Sites

IEEE AESS - St. Mary's - Survivability and Combat Tactics in StarCraft II

#multi-agent #reinforcement #learning #specification #gaming #StarCraft #II

Multi-agent reinforcement learning agents can learn odd behaviors and act unexpectedly. This research presentation focuses on how to shape, analyze, and improve agent behavior during skirmishes in the video game environment of StarCraft II. We demonstrate agents can improve survivability with proper rewards, develop of new behaviors based on their enemies, retain combat tactics using transfer learning, and accelerate learning with collaborative expert training.