public class DynamicAabbTree<T> : BasePartition<T>, 
	ISupportClosestPointQueries<T>, ISupportFrustumCulling<T>

Public Class DynamicAabbTree(Of T)
	Inherits BasePartition(Of T)
	Implements ISupportClosestPointQueries(Of T), ISupportFrustumCulling(Of T)

generic<typename T>
public ref class DynamicAabbTree : public BasePartition<T>, 
	ISupportClosestPointQueries<T>, ISupportFrustumCulling<T>

type DynamicAabbTree<'T> =  
    class
        inherit BasePartition<'T>
        interface ISupportClosestPointQueries<'T>
        interface ISupportFrustumCulling<'T>
    end

Type Parameters

T

The type of item in the spatial partition.

The DynamicAabbTreeT was inspired by the dynamic bounding volume tree (btDbvt) as implemented in the Bullet Continuous Collision Detection and Physics Library. (Original btDbvt implementation by Nathanael Presson.)

The DynamicAabbTreeT was designed to manage deformable objects efficiently. It should be uses for CompositeShapes or TriangleMeshShapes when the the contained shapes or triangles are updated at runtime.

It can also be used as a collision detection broad-phase (see BroadPhase).

Incremental Optimization: When items in the AABB tree are added, removed or moved the AABB tree might become unbalanced and less optimal for collision detection. Therefore, the DynamicAabbTreeT tries to optimize its tree structure over time. In each frame (time step) it performs a number of optimization passes. The amount of optimization per frame can be controlled by setting OptimizationPerFrame.

Motion Prediction: The dynamic AABB tree is further optimized for models/space where items are constantly moving. The DynamicAabbTreeT automatically detects when items are moving. It adds a small margin (see RelativeMargin) to the AABB of these items to account for small random movements ("jittering") and it extends the AABB in the direction the items are moving (see MotionPrediction). This reduces the number of required tree updates per frame. However, the downside is that the AABB tree is more conservative (safe, but less accurate). It might return more overlaps ("false positives") than other types of spatial partitions.

Motion predication can be enabled by setting EnableMotionPrediction. The feature is disabled by default.

AdaptiveAabbTreeT vs. DynamicAabbTreeT: The AdaptiveAabbTreeT and the DynamicAabbTreeT are similar data structures. As a general rule, the AdaptiveAabbTreeT should be used if

• The entire spatial partition is invalidated regularly by calling Invalidate.
• The spatial partition is used to compute inter-object overlaps and intra-object overlaps (see EnableSelfOverlaps) are disabled.

• Individual items are invalidated frequently by calling Invalidate(T).
• Items are added or removed frequently. (Inserting or removing individual items into/from a DynamicAabbTreeT is fast.)

The AdaptiveAabbTreeT should not be used as the collision detection broad-phase (see BroadPhase). Whereas, a DynamicAabbTreeT can be used as the collision detection broad-phase.

However, please note these are just general rules. You should always try different ISpatialPartitionT types and measure which one yields the best performance in your application.

Special handling of self-overlaps in GetOverlaps(ISpatialPartitionT): If GetOverlaps(ISpatialPartitionT) is used to test an AABB tree against itself then overlaps of an item with itself are not returned. That means, each item A overlaps with itself, but (A, A) is not returned. And if two different items overlap, only one overlap is returned, for example: If item A and item B overlap (A, B) or (B, A) is returned, but not both.

Reference