And now it’s getting interesting: Let’s look into how we can animate our models! The by far most popular method for animating rigid bodies in real-time is called skeletal animation, mostly due to its simplicity and run-time efficiency. As the name already suggests: We associate our model with a skeleton and perform all deformations based on this skeleton. Notice the similarity to human beings: our motions are constrained and guided by a skeleton as well.

Taking a closer look at the human skeleton, we can identify two major parts: joints and bones. Bones are rigid structures that don’t deform: Put too much pressure on a bone and it will break – something almost every one of us has experienced already :(. Joints in turn are dynamic: they can rotate with various degrees of freedom (e.g. shoulder joint vs. elbow joint). Each bone is attached to at least one joint and thus a rotation of the joint will cause a rotation of the attached bones. If you rotate your elbow joint for example you’ll see that the directly attached bones (Radius and Ulna, which connect the elbow joint with the wrist joint) are rotated around the elbow joint’s local rotation axis. But also the indirectly attached bones like in your fingers and wrist are rotated around the same axis as well. Our skeleton thus defines a hierarchy where the rotation of one joint will also rotate all joints (and bones) in the hierarchy ‘below’. Lets illustrate this with a simple example: Take a simple cylinder (1). We create a skeleton consisting of 4 bones and 5 joints and attach it to the cylinder (2). The skeleton hierarchy is simple: each bone has one child (3). Now let’s rotate joint 2 by a couple of degrees. All joints below joint 2 will rotate around the local rotation axis defined by joint 2, resulting in the deformed cylinder as shown in (4).

So let’s recap what we’ve gathered so far:

- A skeleton represents a hierarchy of joints and bones
- Each joint can rotate around a local rotation axis
- The rotation of a given joint will cause rotation of all joints in the hierarchy below
- The mesh is bound to the hierarchy such that it will deform with it.

Now how can we model this mathematically? Let’s look at our wrist joint again and remember that it is connected to the elbow joint via the Radius bone. Consider a point in the local coordinate system of the wrist joint: We can express it’s position relative to the local coordinate system of the elbow joint by rotating it around the wrist joint and then translating it along the Radius bone:

Going one step up the hierarchy, we can express it’s position relative to the local coordinate system of the shoulder joint by rotating around the elbow joint and translating it along the Humerus bone:

Inserting into the formula for yields

which simply corresponds to a concatenation of the transforms for the wrist joint and the elbow joint. This example shows that if we express each joint’s transform in the local coordinate system of it’s parent, it is extremely simple to transform a point local to one joint into world space: all we need to do is concatenate the joint’s transform with the transforms of it’s predecessors in the skeleton hierarchy and transform the point by the result.

Formally, defining the rotation and translation transform of joint as and the concatenation of two transforms as we can write the world space transform of joint as

for the sequence of parents of joint . This gives us a simple method of computing the world space transforms of all joints in the skeleton. In order to deform the skeleton over time, all we need to do is animate the local transforms of the joints. This is usually done via keyframe animation, where we store a sequence of deformation parameters over time. For example, if we want to animate the bend of the elbow joint we’d store the sequence of rotation values for the elbow joint along with the corresponding animation time as a sequence of (rotation,time) values. At runtime we’d search the sequence for the rotation value that best matches the current animation time and then change the rotation of to represent that value.

Please note that we’ve implicitly encoded the rotation of a given joint and the translation of it’s corresponding bone into one transform. Thus, in the so defined skeleton we don’t need to distinguish between bones and joints anymore and we will use the term bone and joint interchangeably.

Let’s look at some code now: I defined a character’s skeleton as a simple class that stores the list of joint transforms as a stl vector of 4×4 Matrices. In order to represent the hierarchy structure, I store the index of each joint’s parent as integer value in another stl vector:

class Skeleton { std::vector<aiMatrix4x4> mParents; std::vector<int> mTransforms; â€¦ };

In this convention we can find the transform of a given joint in `mTransforms[i]`

and it’s parent in `mParents[i]`

. If a given joint doesn’t have a parent I set `mParents[i] = -1;`

. As discussed above, we can now simply compute a joint’s world transform by concatenating it’s local transform with the transforms of all parents:

aiMatrix4x4 Skeleton::getWorldTransform( int bone ) const { int p = mParents[bone]; aiMatrix4x4 result = mTransforms[bone]; while( p <= 0 ) { result = mTransforms[p] * result; p = mParents[p]; } return result; }

Well written , short and educational . Thanks for the aritcle and links .