Ragdoll rendering: skeletal mesh deformation from bone transforms #29

New issue

Open

opened 2026-03-18 09:30:31 +00:00 by kit · 0 comments

kit commented

2026-03-18 09:30:31 +00:00

Owner

Currently entities using MDL models are rendered as rigid meshes in their rest pose (T-pose). Bone transforms are baked into vertex positions at load time (mdl/loader.js computeBoneWorldTransforms), and the entire mesh moves as a single rigid body based on entity position/rotation.

For ragdolls (and eventually animated entities), we need to deform the mesh using per-bone transforms.

Server side

Read per-bone transforms (position + quaternion) from ragdoll entities via the engine
Include bone data in the entity snapshot binary format
Only needs to apply to entities with active ragdoll physics (not all entities)

Client side

Switch from baked rest-pose vertices to GPU skinning (bone matrices as uniforms or a texture)
Build the mesh with bone indices and weights preserved from VVD data
Apply per-bone transforms each frame instead of a single rigid body transform

Notes

VVD files already contain bone indices and weights per vertex — we just discard the skinning relationship during the bake step
Source engine ragdolls use vphysics bone simulation, so the bone transforms come from the physics system rather than animation sequences
This is a prerequisite for proper animation support later (same skinning pipeline, different source of bone transforms)

Currently entities using MDL models are rendered as rigid meshes in their rest pose (T-pose). Bone transforms are baked into vertex positions at load time (`mdl/loader.js` `computeBoneWorldTransforms`), and the entire mesh moves as a single rigid body based on entity position/rotation. For ragdolls (and eventually animated entities), we need to deform the mesh using per-bone transforms. ### Server side - Read per-bone transforms (position + quaternion) from ragdoll entities via the engine - Include bone data in the entity snapshot binary format - Only needs to apply to entities with active ragdoll physics (not all entities) ### Client side - Switch from baked rest-pose vertices to GPU skinning (bone matrices as uniforms or a texture) - Build the mesh with bone indices and weights preserved from VVD data - Apply per-bone transforms each frame instead of a single rigid body transform ### Notes - VVD files already contain bone indices and weights per vertex — we just discard the skinning relationship during the bake step - Source engine ragdolls use `vphysics` bone simulation, so the bone transforms come from the physics system rather than animation sequences - This is a prerequisite for proper animation support later (same skinning pipeline, different source of bone transforms)