When you animate a computer graphic character by hand, you have complete control over everything. The placement of the feet, the head, the hips and shoulders, all of the subtle indicators of mood and intent, are under your explicit control.
Of course it can be a tedious process, but it is also a very rewarding one. You are essentially being the actor in a performance, but rather than performing with your own physical body, you are performing within a body by proxy.
But when you are making interactive content, this approach to character animation can start to break down. An interactive character needs to respond to unexpected events, turning or looking or reaching in response to something that could not have been planned for beforehand.
That’s why procedural methods can be so useful. They are designed, from the ground up, to be able to respond to unexpected changes in the environment. But nothing comes for free. In order to provide that power, these methods require the animator to let go of the precise control afforded by hand animation.
How, exactly, does procedural animation make it possible for a character to respond to unexpected events? The key idea is that within every procedural animation system is a model of behavior. Whereas in traditional animation there is no such model — which means that any unexpected change will start to break things.
For example, if you blend together two hand-crafted animations of a character walking, the result will likely be a character whose feet begin to slide along the ground. That’s because, whereas the animator knows avoid sliding feet when making each individual walk animation, there is no knowledge in the computer that such a thing is important.
So any change in the animation, like blending it together with another animation, is likely to create that sortsof weird results. The computer doesn’t care about things like foot sliding, but viewers of the animation care a lot. As soon as an animated character starts to move in a way that is weird and impossible, viewers stop believing in that character.
In contrast, procedural animations have such constraints built into them. In a properly designed procedural animation system, it is literally impossible for a character’s feet to slide along the floor, unless the animator specifically directs the character to do so.
When is this power relevant? Well, if you’re making a movie it is not relevant at all. You’re pretty much always better of hand animating, since everything in the character’s environment is known beforehand.
If you’re creating a computer game, it may be relevant, or maybe not. You still have a lot of control over how the player sees the character, which gives you cover to hide a lot of sins in blended animation.
But if you are creating an interactive character in virtual, augmented or mixed reality, the balance changes. In those media, an observer can view an animated character from any position, and is generally free to move her head to view the character from a different angle.
In these newer media, any undue repetitive movement, and any artifacts caused by bad motion blending, will jump right out, and will quickly dominate the visual experience. That’s why I think that procedural animation is going to become ever more important in the next few years, as interactive animated characters move out of screens and into the world around us.
I agree, but I don’t expect procedural to be a panacea. The model that drives the procedural animation still needs to anticipate all sorts of situations, which can be a daunting task.
This is one of the well-known problems in the very different world of interactive fiction (text adventures). There’s an endless debate over how much the program should be a general purpose simulation of the environment (procedural) versus a carefully crafted illusion of a consistent environment (manual).
It turns out both are very hard to do well and nearly impossible to test exhaustively, even with automation. The player has a wide range of options and the world has vast number of possible states–the test matrices are intractable.
Consider a puzzle in which the player must cut a rope. The author expected the player to slice it with the sword. For whatever reason, the player doesn’t have the sword so instead smashes the priceless vase and tries to use one of the sharp fragments as a knife to saw through the rope.
Games in the manual style will fail unless the author had anticipated using the shards as knives. But games in the simulation style (procedural) might also fail because the author had failed to tag the shards with the “sharp” attribute that the simulation relies on to see if the tool can cut the rope.