Game Engine Architecture

580 11. Animation Systems

something with her shoulders, head, and eyes and making a gesture with her

hands and arms. The movements of diff erent parts of the body aren’t gener-

ally in perfect sync—certain parts of the body tend to “lead” the movements

of other parts (e.g., the head leads a turn, followed by the shoulders, the hips,

and fi nally the legs). In traditional animation, this well-known technique is

known as anticipation [44].

This kind of movement seems to be at odds with a state-machine-based

approach to animation. Aft er all, we can only be in one state at a time. So how

can we get diff erent parts of the body to operate independently? One solution

to this problem is to introduce the concept of state layers. Each layer can be

in only one state at a time, but the layers are temporally independent of one

another. The fi nal pose of the skeleton is calculated by evaluating the blend

trees on each of the n layers, thus generating n skeletal poses, and then blend-

ing these poses together in a predefi ned manner. This is illustrated in Fig-

ure 11.58.

The Uncharted engine uses a layered state architecture. The layers form

a stack, with the bott om-most layer (called the base layer) always producing

a full-body skeletal pose and each upper layer blending in a new full-body,

partial-skeleton, or additive pose on top of the base pose. Two kinds of layers

are supported: LERP and additive. A LERP layer blends its output pose with

the pose generated by the layer(s) below it. An additive layer assumes that its

output pose is always a diff erence pose and uses additive blending to combine

it with the pose generated by the layer(s) below it. In eff ect, a layered state ma-

Base Layer

State A State B State C

Variation Layer (Additive)

D E G

Gesture Layer (Additive)

H I

Gesture Layer (LERP)

J K

F

Time (τ)

Figure 11.58. A layered animation state machine, showing how each layer’s state transitions

are temporally independent.