Collision System Misuse
The collision system can detect most types of collisions very rapidly. However, it is possible to construct situations that the collision detection system just can’t handle. This page lists a few of these things, and how to improve them.
Colliding with Visible Geometry
Panda3D supports calculating collisions against visible scene geometry. However,
this method is not at all efficient. It is far more efficient to construct
specific Collision Solids to represent the various objects in the scene.
Even collision geometry constructed from
CollisionPolygon objects is
significantly more efficient than colliding against the renderable geometry!
Various exporters are able to automatically generate optimized collision geometry during the export process. Consult the documentation for your exporter to find out how to do this.
When you have constructed the optimized collision geometry, don’t forget to disable checking for collision against the visible geometry, by setting the Collision Bitmasks appropriately.
Lack of Hierarchy
A naive approach for exporting collisions for a large scene is just to put all
of the collision solids under a single
CollisionNode, or perhaps many
CollisionNode objects under a single parent node.
This is very inefficient for a large number of solids. Panda3D will need to test for collisions against every single object in a row.
It is far more efficient to structure your collision scene in the form of a quadtree or octree, or even a binary tree, grouping close collision nodes together under a common parent node. This allows Panda3D to test against the parent node, and only proceed to test the child nodes if this passes.
If the collision geometry is constructed from the version of the mesh that is used for rendering, it is possible that there are so many triangles in the mesh that it is slow to test against. In these situations, it is recommended to instead construct a simplified version of the mesh. Ideally, this simplified mesh consists of Collision Solids such as spheres, boxes and capsules, but in some situations it is necessary to build it up from individual polygons. A low-poly version of the mesh can then be used instead, either by constructing it manually or by using an automatic quality reduction algorithm (such as the Decimate modifier in Blender) to create a collision mesh with fewer polygons.