Generating Real-Time Strategy Game Units Using Search-Based Procedural Content Generation and Monte Carlo Tree Search

RaidEnv: Exploring New Challenges in Automated Content Balancing for Boss Raid Games

“…Pokémon [24] Card Deck Building High-performance Automatic Team Assembly microRTS [25], [26] Skill, Stats High-performance Multicontent Balancing (Skill, Stats)…”

“…The PCG criterion describes the content types that are employed for content generation. Several game variables with character properties (e.g., skill and stats) [9], [25], [26], map layout [8], [10], [19], [21], [23], card decks [20], [24], and game rules [7], [8] have been investigated. The game rules and character properties are related to the game mechanics and map layout, while the card deck is considered playable content.…”

“…The multi-agent criterion describes the most frequently addressed problem definition in the game environment. The highperformance class [13]- [15], [19], [24]- [26] represents the environment that addresses the general multi-agent problem and the cooperative gameplay, while ad-hoc and environmental generalization are specific sub-problems addressed in the specific environment. Ad-hoc generalization [11], [12], [18] is the generation ability of gameplaying, where unseen agents are not considered in the training step; AI-human coordination [23] refers to the agent's ability to adapt to human behavior policy, and environmental generalization [16], [17] refers to the generalization ability to control diverse types of characters using a single policy.…”

“…In single-content balancing, the balancer module regulates single-content item (e.g., skill) to meet game results for a particular setting, while in multi-content balancing, the balancer regulates two or more content elements (e.g., skill and stats). Evolutionary methods are employed in [26] to balance units (characters) in real-time strategy (RTS) games. In particular, [26] simulates the game with play-testing agents of varying proficiency levels.…”

“…Evolutionary methods are employed in [26] to balance units (characters) in real-time strategy (RTS) games. In particular, [26] simulates the game with play-testing agents of varying proficiency levels. The proposed environment, i.e., RaidEnv, can support the multicontent balancing of the character's skill and stats; however, only the skill content item was considered in the current study.…”