"Why Halo Infinite's Bots Play More Like Humans | AI and Games #71" is a video published by AI and Games, which discusses how 343 Industries, the developers of Halo Infinite, created AI-controlled bots that exhibit human-like behavior in the game's multiplayer modes. The video features insights from Sarah Stern, the senior multiplayer designer for Halo Infinite, and Chin Direly, a senior gameplay engineering lead on the Halo Infinite multiplayer team.
Halo Infinite is the first entry in the series to offer a free-to-play multiplayer mode, attracting a diverse range of players with varying skill levels and experiences. To ensure a positive experience for all players, the developers designed AI-controlled bots to act human-like in their behavior, using traditional AI techniques and a different approach to behavior design.
The AI-controlled bots in Halo Infinite are built using behavior trees, a common mechanism for designing and implementing behaviors for non-player characters in games. The bots exhibit many common behavioral traits that players would expect from a human player, such as engaging in combat, picking up objectives, interacting with objects, and traversing the gameplay space.
To make the bots appear more human, three main design pillars were considered: individual skills, a utility AI system, and a nuanced understanding of the game mode. The individual skills of the bots, such as strafing, aiming, grenade usage, melee combat, and confidence in combat, were designed to reflect human behavior and gradually improve as players get better at the game. The utility AI system, which runs inside the behavior trees, conducts a thorough analysis of the current game state to determine how a given bot should react, ensuring changes in priority reflect human playstyles. Lastly, the bots have a more nuanced understanding of the game mode, factoring in objectives and other game-specific elements into their decision-making.
To achieve realistic bot behavior, the developers identified five key skills that reflect the Halo combat dance and assigned levels of competence to each skill, ranging from Level 1 to Level 4. These levels were then mapped to the four difficulty levels of the bots: recruit, marine, ODST, and Spartan. The bot team analyzed recorded footage of players and observed colleagues at 343 Industries to define these levels and ensure the bots' behavior reflected actual human play.
The utility AI system in Halo Infinite uses a technique known as "ambition" to prioritize behaviors based on the current game state. Ambitions are assigned a utility value between 0 and 1, with higher values indicating greater importance. The bots' confidence value, which determines whether they engage in a fight or retreat, is calculated using this utility system.
To avoid common issues with utility-driven AI systems, such as constantly switching between actions, the developers added a short boost to the utility value of a chosen action, ensuring the bot commits to a decision for a short period of time. Additionally, on higher difficulties, the bots can share information with each other, such as map callouts and last known positions of enemy players.
Despite concerns about the speed of the Lua programming language used for the utility calculations, the developers found that it was sufficient for the bots to process information within the desired time window, which is around 200 to 250 milliseconds for an average human being.
In conclusion, the AI-controlled bots in Halo Infinite were designed to exhibit human-like behavior, providing a more engaging and realistic experience for players of all skill levels. By focusing on the core skills that human players develop and replicating them in the AI-controlled Spartans, the developers have created a valuable tool for helping new players learn the ropes and improve their skills in the game.
