Ammo.js Heightfield vs Static Triangle Mesh Performance

This article compares the performance differences between btHeightfieldTerrainShape and static triangle meshes (btBvhTriangleMeshShape) in ammo.js. When building physics for 3D web applications, choosing the right shape for terrain and environment collision is critical for maintaining high frame rates. We will examine memory consumption, CPU collision detection efficiency, initialization overhead, and functional limitations to help you make an informed architectural decision.

Memory Consumption

The btHeightfieldTerrainShape is highly optimized for memory efficiency. It represents terrain using a regular grid of elevation data, storing only a single height value per grid intersection (typically as a flat array of floats or integers).

In contrast, a static triangle mesh (btBvhTriangleMeshShape) requires storing three-dimensional coordinate data for every vertex, an index buffer to define the triangles, and a Bounding Volume Hierarchy (BVH) tree structure to accelerate collision queries. For large-scale maps, a static triangle mesh can easily consume several times more memory than a heightfield of equivalent resolution, which can quickly lead to browser tab crashes on mobile devices with limited RAM.

Collision Detection Speed

During the physics simulation, ammo.js must perform collision detection between dynamic rigid bodies and the terrain.

• Heightfields: Because the data is structured as a regular 2D grid, the physics engine can project a colliding body’s bounding box directly onto the grid. This allows for \(O(1)\) constant-time lookup to locate the exact terrain cells involved in the collision, leading to extremely fast narrow-phase collision calculations.
• Static Triangle Meshes: To find collisions, the engine must traverse the AABB (Axis-Aligned Bounding Box) tree structure of the BVH. This traversal operates at \(O(\log N)\) logarithmic time relative to the number of triangles. While highly optimized, tree traversal still introduces more CPU overhead and cache misses than direct grid indexing, resulting in slower collision resolution as the complexity of the scene increases.

Initialization and Loading Times

Initializing a btBvhTriangleMeshShape requires ammo.js to build the BVH tree out of your mesh data. For detailed meshes with tens of thousands of triangles, this calculation is computationally expensive and runs on the main browser thread (unless offloaded to a Web Worker). This can cause noticeable stuttering, frame drops, or long loading screens during scene initialization.

Creating a btHeightfieldTerrainShape is nearly instantaneous. The engine simply references the existing array of height values, requiring no complex tree construction during startup.

Functional Differences and When to Use Which

While btHeightfieldTerrainShape is superior in performance, it has strict geometric limitations: * Heightfield Limitations: It only supports 2.5D surfaces. It cannot represent overhangs, caves, tunnels, or vertical walls that overlap on the vertical axis. * Static Mesh Flexibility: A static triangle mesh supports arbitrary 3D geometry, making it necessary for complex environments containing arches, indoor structures, cliffs, and vertical obstacles.

Summary

For standard outdoor terrains, btHeightfieldTerrainShape is the superior choice, offering drastically faster initialization, lower memory usage, and better CPU performance. Use btBvhTriangleMeshShape only for complex, arbitrary static geometry like buildings, caves, or indoor levels where overlapping 3D structures are functionally required.