Captured by Motion
A great many things comprise the atmosphere players enjoyed in Pathologic of 2005. The sick, foggy town. The dilapidated structures sprayed here and there. The odonghe and herb brides and kids, these little stray dogs roaming the streets. The languid townsfolk, suspicious of a stranger yet seemingly nonchalant about the odonghe and feral kids.
Of many things Pathologic was, it was never a beautifully animated game.
Don’t get us wrong, there were interesting choices in how some of the townspeople moved. Drunks, determined to move in their swaying plod to their last day are just one example. Kids also had a few lovely, albeit unpolished motions. Moreover, there was a lot of it.
We shall remind you in case you have forgotten
Since there was little hope the amount of animation was going to get smaller, it was natural for us to turn to motion capture—a technology that had long earned it place in the game development industry.
Two Ways of Doing It
In modern game development, two primary approaches to making character animation are used: keyframe animation and motion capture.
To make keyframe animation, an animator manually moves the character’s body parts to create the required poses. The in-betweens are automatically calculated by smoothly moving everything from one pose to the next. Animation in The Void was all keyframe. The desired level of quality and the time you have will affect how much detail the animation gets.
The term comes from hand-drawn animation where studios produce principal drawings based on storyboards, then the animators draw the key frames and important intermediary frames to outline the action. Lastly, the frames between those frames, the inbetweens, are drawn by assistant animators (quite a tedious job!). This is how pose to pose animation works.
Motion capture essentially lets you record movement from an actor: you make them wear a specially designed suit that allows tracking the movement and then perform all the necessary actions. We used marker motion capture, i.e. the suit had small balls attached at every important body part.
Multiple cameras are installed inside the room where the shooting takes place. The computer analyses what it sees and, with some noise, tells you the coordinates of each marker. Finally, the motion of a dummy skeleton is reconstructed based on this data.
As is often the case with dummies, it does dumb things sometimes.
Amongst animators, it is common knowledge that motion capture is not animation—even though you can make animation from motion capture. It’s just a way of saving time.
Done wrong, it is at times jerky and may look generic or plain wrong. But done right, it allows the creative leads more flexibility: if what you have recorded looks nothing like what you actually wanted, you can immediately redo the whole thing. Which is why, just like in movies, you always shoot extra takes—for the animator to sort out later.
Even with extra takes, it ends up being way faster than doing everything by hand.
However, unlike keyframe animation, motion capture is not guaranteed to provide you with clean data. The recording equipment only knows what it sees through the cameras—i.e. the markers, the small balls attached to your actor. Markers occasionally get “lost” when visible through none of the cameras. You are limited by the resolution, too; if the actor only moves slightly, there is always a little jiggle from markers moving back and forth between adjacent pixels.
Just so that we’re on the same page, here is how raw motion capture looks sometimes—unless the studio has dozens of cameras.
And then the animator’s job is fixing each dislocated joint manually to turn the whole thing into something beautiful. Or at least humanoid.
At least presumably.
But the fun doesn’t end here.
See, making a unique, movie-like animation is one thing—but not the most essential. In video games, most animations—even the long and complex ones like the dance of a Herb Bride below—have to loop back, so that they can be played infinitely.
Performed by Elena Ermak, the Herb Bride’s dance is long and intricate—and, being a real dance, it’s hard to make a loop out of. In hindsight, telling Elena that we needed her to return to her starting position during the actual dance would probably have helped too.
Actually, unless the talent is a robot, they never return to the starting point or repeat their motion perfectly, but usually fixing this is simple. This dance, however, had no repetition at all.
This is where the animator comes in again, inventing and then constructing a loop that was not there at first. We scanned the whole thing for motions that looked about the same for a second or two—and happened at about the same place. Preferably, they had to have a decent chunk of movement between them for the dance to end up long.
After finding such motions, we needed to match them perfectly; the beginning and the end of the dance were still over two feet apart. In the end, we managed to hide this mismatch in the girl’s steps, allowing her to slide slowly into the required position. The only thing left was to fix the sliding feet; the latter is rather trivial.
The result is what you see in the game.
Patience, diligence, and talent create brilliant expressive animations. However, patience, diligence, and talent also take an awful lot of time. Resource allocation can easily become an issue even for a bigger developer—and is absolutely crucial to an indie like us.
Take something as simple as walking, for example. Nothing flashy, just a single person being able to pass you by in the street—which you are unlikely to even register (and if you do, it is usually not in a good way). How many separate animations does this passer-by need?
Well, they have to be able to stand still, walk, and run. To start walking and stop doing so. Crucially, they also need to learn to turn in place and start walking in different directions. The exact number of animations required may vary. For example, Assassin’s Creed’s movement system, including all the climbing and parkour, had several hundred animations (over 500 in latter installments). Our numbers, of course, are much humbler: we started with about 25 to 30 for just basic movement.
Unless you do motion matching to find animation automatically all these animations should match and flow into one another really well—which is, once again, more manual labour for the animators; more patience, diligence, and talent; more time. And that’s just for a single type of character—which we wanted to have more of.
Fortunately, this is where the medium comes to our rescue. You see, just like everything else in a game an animation is, in the end, just a collection of 0s and 1s—that can be manipulated mathematically. Calculating an automatic “mix” of two animations or blending, has been in games for quite a while. Say, if you need your character to turn 70° left, you can actually shoot 45 degree and 90 degree turns and then mix them to get an intermediate angle. The end result is pretty much the same as the real thing (provided that both motions have similar dynamic and timing).
So this is what we did. We cheated by first creating the honest-to-god animations of turning 60°, 120°, and 180° left and right—due to the angles chosen, we only needed 6 unique animations. Then the game engine does the rest.
In the end, choosing 60-degree steps over 45-degree ones saved us 4 unique animations for each character type—a seemingly mundane victory that has a huge impact on the overall quality of the game, since each trivial animation saved brings us one step closer to creating an additional exquisite Herb Bride Dance—or something similarly complex.
It’s true that the face of the game is heavily defined by the artists, actors, and modellers. But one has to keep in mind that in the end all they do is make and nudge the puppets.
And, fittingly for Pathologic, in a 3D game it’s the puppeteers who always have the final word.