Motion Capture Or Animation For Games?

Thanks to TotalGames.net for their article discussing whether videogames should use traditional animation or motion capture to capture the movements of in-game characters. The piece points out: "One of the major problems with motion capture is the way that moves can sometimes appear disjointed and separated, as a character goes from one set of moves to another", but an advocate for motion capture comments that the process is "..a lot faster, as long as you can retain the subtleties from the point of motion capture to the raw data to the point where it reaches the engine." Can you tell the difference?

Comments

Motion capture

[recursiv]

Hand animation never seems to capture the feeling of "weightiness" properly. The models just jerk around as if they had no mass or center of gravity. With motion capture, the center of gravity is always apparent, and the model doesn't seem like a hollow marrionette.

Re:Motion capture

[neostorm]

If the animator is any good, the sense of gravity and weight is much better in models animated by hand, rather than motion capture.
Motion capture will always look more realistic, but so far there is a severe lack of animation skills in the industry. Mark of Kri and Jak and Daxter (both for PS2) are the only games I can think of off the top of my head that have well done character animation by hand.

In the end it comes down the type of game you're creating. You would never give a colorfully animated character a motion captured animation set, and giving realistic people exaggerated animations would give the game an entirely different tone.

Re:Motion capture

[NanoGator]

"Hand animation never seems to capture the feeling of "weightiness" properly. The models just jerk around as if they had no mass or center of gravity. With motion capture, the center of gravity is always apparent, and the model doesn't seem like a hollow marrionette. "

Well you've been a bit vague here. Possible reasons could be:

- Animators just getting started
- Technical limitations such as limited RAM. (Very possible.)
- Too little time.
- Motions could have been intentionally simple because when you move the mouse, the character moves according to some equation. Ever notice that when you rotate a character in Quake, his feet just rotate like he's on a pedestal?
- Intentional design. This may sound funny but maybe they didn't want the character bobbing around if you're trying to snipe it.
- Bad implementation. Animation is very complex. Getting from the artist to the programer can be nasty, especially when unexpected nuances appear.
- It wasn't an animator at all, somebody just trying to make the model move.

I could keep going but without a game in particular, I cannot give you an answer why you feel that way. I do think you're unfairly generalizing, though.

Re:Motion capture

[Goldberg's Pants]

When you fell, you fell like a 600 pound man:)

"Another World" did the motion capture thing right. Also "Flashback" (both done by the same company) had good motion capture too.

There's a place for both in games I think. Motion capture is great, and a must for 3D games, but hand animated sprites still kick all kinds of ass in the right games.

Re:Motion capture

[Goldberg's Pants]

"Out of this World" was it's other title. Screenshots here.

Animation v motion capture

[Xerxes of Zealot]

It depends on the game, am I right?

I dont want to play Mario and watch some actor whos been digitally captured, I wanna see some goddamn animation.

Likewise, I don't want to play Manhunt and see animation, then I want motion capture...

Re:Animation v motion capture

[hackwrench]

You're assuming you can't wrap animation around motion capture data.

Animator's retort

[MobyDisk]

This guy is the animation advocate. His arguments are less than impressive:

...computer processing...takes out the creative flair that brings a character to life, thus making characters maybe more lifelike but somewhat lacking in life.

Well, that clarifies it. More lifelike, but lacking in life.

Obviously, animating traditionally takes longer than motion capture but you do have to spend a lot of time processing the data...So in a lot of cases a skilled animator...will achieve the same if not better results in not much more time than if the moves were motion captured. Motion capture is probably quicker....

So it takes longer, but in some cases it maybe traditionally isn't longer, but it's probably not quicker. Other great quotes:

Motion capture cannot always capture the sheer fluidity that many athletes possess and as a result this supposed realism can sometimes undermine all the motion capture artists' hard work.

Actually, that is the entire point of motion capture: It captues the fluidity of the actor exactly.

Re:Animator's retort

[NanoGator]

"Actually, that is the entire point of motion capture: It captues the fluidity of the actor exactly."

Eh not exactly. Mocap has two problems.

1.) Capture's not always perfect. There are glitches. Bad tracking. Etc. You get a very noisy graph for each channel, and they are prone to error. Imagine a sound wave. If you take a tape recorder and record a solid tone, you'll see that the tone doesn't generate a perfect wave. Lots of tiny little zig-zags there. The more 'perfect' the mic and capture equipment is, the less visible those zig-zags are. Those variances from what the value should be are noise. Crappy mic = very jagged. Kick ass mic = very smooth. Mo-cap is very much like that. Actually the motion graphs resemble a sound in some ways. If you have a ball that bounces up and down, the motion is real simple. It, in an over-simplistic sense, is a logarithmic curve. But with mo-cap, you have all these cameras watching a sensor strapped to a ball watching its motion. There are inaccuracies there. Careful inspection will reveal tiny errors in the motion. If the sensor is obscured from view, then the computer has more work to do.

That sort of make sense? In a round-about way, I'm saying that capturing mo-cap is sort of like capturing sound. It can be error prone.

2.) Athletes operate in uniforms that are as least restrictive as possible. This is in stark contrast to what a mo-cap person has to wear. They have to mind all these little sensors. They have to stay in one little spot for all the cameras to watch. Etc. It can be a hinderance.

Sorry I haven't read the article, but I think from that quote that's what he was after. Even if I got the context wrong, it is interesting to know. Frankly, I'm scared to work with mocap data. In the ball-bouncing example, I don't have 3 keyframes to play with. I have one for every single frame. (Unless the software is nice enough to find a curve for me...)

Re:Animator's retort

[clambake]

Actually, that is the entire point of motion capture: It captues the fluidity of the actor exactly.

Actually, I see this as one of the main problems. Since the actors aren't really hurtling off 600 foot towers or doging real mini-missles fired by real mech commandos, the movements seem "wrong" somehow. They are all done in a green room with little reflective ping-pong balls which makes even the best actor lose some of his natural motivation. Without the set and scenery, he has a hard time "acting natural" and tends to overact, mime-like, to emphasize the things that are supposedly around him, but that he cannot see. So once the special effects are added in, he's making these big unnaturally sweeping gestures to pick up a pencil when an animator would never have done that, so the animation looks more "real" than the real life guy...

Re:Animator's retort

[SuiteSisterMary]

Depends on what you're doing, actually.

With Rainbox Six, for example, they took a man, punched him *really hard* in various spots with a pole, and recorded him falling down. That simulates being shot.

The problem with this, or course, you see in games like Splinter Cell; Sam's running, or jumping, or rolling, or shooting. But he can't really run into a jump, miss, land, roll into a shooting stance and start peeling off rounds.

But nowadays, with tons of processor, and a good physics model, it can be just as good to build your models properly, and let the engine figure it out. Terra Nova: Strike Force Centauri, Die By The Sword (I believe) and Oni, for example, played with this early on; Unreal Tournement 2003 seems to be the current champ.

Myself, for video games, I'd avocate using mocap to more accurately build your models and tweak your physics engine, then let the engine actually do things. When my model's running up a hill, I want to see him leaning forward to place his center of gravity. When he gets shot, I want to see him fall back, then down, then start rolling down the hill. When he's lying on the ground, I want to see his arm tracking where I'm aiming; I want to see him use his other arm to lift himself up off the ground.

This is not an absolute

[NanoGator]

If you can tell the difference either way, then the method was not implemented properly. There is no "is one better than the other?" there's only "which is best for the job."

Consider this: If you want to use mocap, but it causes wiggly meshes, then you have an implementation problem. Do you write the code or adjust the model/actor to solve it, or do you use the video as reference and manually keyframe it? Do you even care? Will the gamer care? Is it worth the money?

I realize this isn't exactly a either/or type of discussion, but it will quickly turn into one. One guy will have seen bad mocap, another will have seen bad keyframing. Done right, either technique will be extremely convincing. It's fun to discuss, but let's keep this out of the realm of absolutism.

State of the Art combines the best of both

[cying]

I strongly urge anyone who wishes to learn more about current research in combining motion capture with animation (similar to image-based rendering techniques applied to the motion domain) to look at these links, one from this year's SIGGRAPH, and a link to several other papers on the topic. SIGGRAPH 2002 had a special track on this (and the bibliography is cited as well).

• Stanford Movement Publications (2002 and before) - especially Kathy Pullen's paper
  
• Motion Synthesis from Annotation
  
• SIGGRAPH 2002 Publications
  

How about a combination of both?

[cloudless.net]

Who says motion capture has to be disjointed and separated? If you insert appropriate animations between different captured actions, the result could look very smooth. Just because many games have crappy motion capture doesn't mean the technology is flawed, we are just waiting for someone to do a better job.

Uh, both?

[GoRK]

No matter which technique you choose, there's going to be some degree of animation involved. To make it look right, most often a hybrid approach can yield the best results. Motion capture is fine and good, but what about when you are trying to model the movement of a 50-ton 20-foot tall rock monster or something? It can't be done accurately with motion capture, but you can use motion capture along with traditional animation techniques to create realistic looking physical movement. Or take gun recoil-- motion capute is probably going to produce something very realistic, but using that data verbatim in a game engine will likely result in poor playability. Animation fills in the gaps.

I don't really see what the argument is about here. There is plenty of work for both the motion capture proponents and the animation proponents!

Re:Uh, both?

[NanoGator]

"Motion capture is fine and good, but what about when you are trying to model the movement of a 50-ton 20-foot tall rock monster or something?"

You use a poor-man's motion capture. You video tape an actor (or a creature?) and anlyze its movements. How many frames does it take a step? What frame does it shift its weight?

Walking with Dinosaurs comes to mind. I watched the making of it, and boy was that cool and insightful. They had these birds that were kind of like pterydactls(sp?) with features similar to bats. They did kind of a four-legged walk with their hands located at the mid-joint of their wings. They had a terrible time animating these guys because they've never seen a creature here on earth that does that. So, eventually they figured out that a guy using crutches walking like a four-legged animal worked quite well. They didn't mo-cap it (that I saw) but they did analyze the vids frame by frame.

Another example popped into mind: Monsters Inc. Got the DVD? Watch the making of it. You can see a guy walking on a treadmill with cardboard tubes hanging from his arms. He was acting like a monster who's arms were really really long.

Yeah I know you weren't quite looking for the answer to this, but what they do to figure these problems out is amazing.

motion blending

[zaad]

There's actually been quite a bit of research to do motion blending so that the transition between states are not noticeably unnatural.

So the real answer is, it's not a limitation of mocap, but current application of the technology.

false choice, perhaps

[cyranose]

As has been pointed out elsewhere, there's a lot of good research on how to combine both. However, the key issues in choosing a method are:

Animated characters move in non-physical ways. A character can turn its head left to right in one or two frames. A human can't (without injury) and that puts a hard limit on mocap's usefulness there (except, see below).

Assuming you want realistic human motions, using a realistic human as model is essential. This can be a living human or a high-quality biomechanical software model driven by an animator or an algorithm. The latter is more interesting, since it allows more than just recombinations of recorded motions.

The main thing missing in motion capture, IMO, is real-time feedback. I worked on a system that used only 12 6DOF body sensors (magnetic, long time ago), but allowed you to drive an animated character in real-time. The effects were really good, IMO, in that the actor could adapt the way a puppeteer learns the motions of her puppet.