The collision detection engine does not account for movement between two frames of animation, a fact that is clearly demonstrated by the Panoramic Camera's ability to accelerate and clip past a level's boundaries in Theater. Halo 3's engine uses a posteriori collision detection. For example, only Brutes can drive Banshees, and only Jackals can act as snipers. Filters Particular tasks are only available to specific types of personnel. Refinements have also been made to that system some are listed below. (where adding s to t would not exceed t’s capacity) The 'greedy approach' to this system, expressed in pseudocode, is: while (S is not empty)įind pair (s,t) that give the minimum H(s,t) for all S x T The only danger in attempting to minimize H( F) is that 'AI can look really stupid with wrong H( F)'. The cost function allows the game to weigh different concerns-for example, the squads 'don't want to travel far' but do 'want to act coordinated' and 'get near the player'. Bungie needed a way to choose between two different distributions. Different tasks and behaviors will be distributed amongst the squad's members. Basically, not every member of a squad will be performing the same tasks or behaviors in a given situation. The cost function ( H( F)) gave Bungie 'a basis for choosing one distribution over another'. The first part of that problem (respecting task-capacity constraints) was referred to as 'bin-packing'. There were two parts to this function: the developers needed to respect all task-capacity constraints and minimize the cost function H( F) - in other words, they needed to avoid assigning too many squads or personnel to a task, and they needed to make the function ( F) as efficient as possible (with H measuring how much it 'cost' to use F). Bungie's goal was to find a mapping F( S)→ T - in other words, Bungie needed a function ( F) that would effectively assign squads to tasks. In mathematical notation, the squads are expressed as a set S with n members the tasks are expressed as a set T with m members. Recurse for children in top priority-group.Collect squads to attempt to distribute.Squads in inactive tasks assigned back up to parent.The specific algorithm has been described as: An example behavior tree can be represented with a multi-level bulleted list: Those nodes' children are specific behaviors. That node's children are the general tasks that the combatant is performing ('hide', 'fight', etc.). The root node of the tree is simply the beginning of a combatant's decision-making process. The behaviors of an individual AI-controlled combatant are also represented as tree structures where each node is a self-describing behavior, such as 'throw grenade' or ' melee'. In addition to specific task trees, there are also three general stages of battle: advance, retreat, and make a last stand. These sub-tasks are assigned priorities ('The most important thing is to guard the door, but if you can, also guard the hallway'), and each is broken into additional sub-tasks ('Guarding the hallway means guarding the front, the middle and the rear of the hallway.'). This task is broken into sub-tasks (such as guarding a main entrance and guarding a hallway). When a task is completed (and turned off), the squad members that were working on it are reassigned to any other tasks that need completion.Īn example is the task of guarding an indoor environment. Individual tasks can 'turn themselves on and off' when a high-priority task is turned on, squad members are reassigned as necessary to ensure its completion. The effect is that individual combatants are assigned to complete the most important tasks first, with leftover personnel being assigned to less important tasks.

Squads are 'poured in at the top' of a tree of tasks, and 'filter down' through that tree. In Halo 3's AI system, squads are given a set of tasks, with each task having priority levels and sub-tasks.

Halo 3 odst drawings