4/26 Swimming Creatures and Soft Body Locomotion

This day we looked at several papers regarding swimming creatures. I thought the idea of modeling both the forces of water on the characters as well as the characters on the water makes sense - there's always two sides to every interaction, and I feel like it might not look correct if you don't model that. The papers that didn't do that didn't have such great results. The characters looked like there was no resistance working against them when they were swimming, like they were just doing the motions in air.

We also looked at a paper on soft body locomotion - characters that bend and stretch and have no skeleton. I thought the way they modeled the muscles in the characters was interesting, but could potentially be limiting. They had good results some of the time, but other examples, such as jumping and rolling, looked extremely bizarre. The motions that were good were really good, but the motions that were bad were really bad. I wonder if that would be solved by adding one more type of muscle, since they only had three: longitudinal, radial, and helical. I don't know what it would be, though.

4/24 Real-Time Deformation

I thought this paper was pretty cool, but I was unsure what type of materials they were trying to create. I thought most of the examples looked fairly realistic, but I was thinking about what objects in real life behave the way those did and I couldn't come up with any. I thought some of them looked a bit like Jello, but if you dropped Jello like those objects were dropped, the Jello would probably break. I also thought about those sticky hand toys:

They seemed a bit like that, but they weren't sticky. I kept thinking and I couldn't come up with any real world examples of objects that behave the way the simulated ones did. Even so, I still found the simulated ones to be convincing. How does that work? I suppose as long as they're physically plausible, they should be believable, but at the same time I think we find things believable if they are similar to things we've seen before. Something to think about.

4/17 Yarn-Based Cloth

This paper immediately caught my attention because I love to knit. Because of that, I understood more than anyone else in the class how yarn behaves, which also, I think, made me more critical. Overall I think their results were pretty good, but I was very disappointed that they didn't make distinctions between different types of yarn or needles. Fingering yarn is different from sock yarn is different from worsted yarn is different from chunky yarn. Wool is different from cotton is different from polyester is different from hemp is different from bamboo. All of these are different with different size needles. The first paper didn't address this at all, it seemed, and the later ones had examples with different yarns but didn't say what they were! As I said in class, the rectangle falling over the ball would be great if it were supposed to be fingering yarn on size 2 or 3 needles, but if that's supposed to be a worsted weight throw blanket it looks way to thin and light. I thought they had pretty great results, but they really needed to give their examples some context.

4/12 Upsampling for Cloth Simulation

I thought this paper was really cool - simplifying cloth simulation so it still looks somewhat believable but doesn't take so long. In discussion in class, I think we were a little bit to critical. No, it doesn't look perfect, so for a movie you're more than welcome to dislike it. But I definitely think if this showed up in a video game, people would be amazed by how great the clothing looks, which is the entire point of the simulation. I definitely think it needs some work, such as fixing the problem of it folding in the same places over and over, but overall I think it's a great idea. Either way it's a vast improvement over what's in video games now. I do think they could have greatly improved the look of their results by making the fabric less shiny though - it made everything seem like silk or satin, even if it was supposed to be a heavy linen.

4/5 Simulating Leaping/Physics of Dance

For the first paper, I was surprised that the knees are barely considered for keeping balance. It makes sense for examples where, say, you're standing still and someone pushes you from behind. In that case, you react mostly by bending forward at the hip, and I'm sure you subconsciously do something with your ankles too. However, the knees are barely used. But what about in the example where the gymnast is trying to stick her landing? She bends her knees a lot there. I suppose you could say that the knees are for the impact and the hips and ankles are for balance, but since it's all one motion I feel like the knees would have to be considered. Maybe it's that I don't entirely understand the physics of it, but I feel like if you're falling or jumping, you use your knees a lot to maintain balance.

As a dancer, I found Physics of Dance to be pretty interesting, too. A lot of the different physical corrections suggested for balance were a bit funny to me, because I don't think about those things at all when I dance. The best solution for balance, I find, is to clench my abs. Of course, I don't really know how that effects things like center of mass, angular momentum, etc.

3/29 Optimizing Walking Controllers

This paper was pretty much an improvement on Simbicon. In order to fix the problem of characters stomping, they added joints in the feet to allow them to curve, as we do due to our toes. It was an improvement, but it still looked bizarre. We mentioned in class though that everyone that tries to add toes keeps their characters seemingly barefoot - I wonder if it wouldn't look so strange if they had the toes move inside shoes. Or even restrict the feet to bend based on the flexibility of most shoes. Either way, the toes were bending too much for too long, creating a really strange motion. Another thing that struck me was how different the results looked when adapted for different body types. The regular, short, and overweight bodies looked awkward, but the tall and thin bodies looked a lot better - believable almost! We haven't see any research that examines how motions differ between people of different body types, but I think that could be useful. It might be that this paper's method is ideal for tall and lean bodies, whereas other equations might be better for short and squat. I think a lot of researchers tend to neglect the many differences in the human body, and treat motions as if they are the same for everyone. Of course, we should perfect one walk before we try to perfect them all, but I do think this is something people should look into.

3/27 Modeling People from Photos

I had to come a half hour late to this lecture, so unfortunately I didn't get too much out of it since I missed all the explanation in the beginning. I did find it interesting that for interpolating between photos, instead of interpolating directly (which can lead to unwanted states in the middle) they interpolate to neutral in between the beginning and ending states. I really liked that they did that - it's such a simple idea, and it makes a huge difference. I think that really shows the answer isn't always a complicated one.

3/22 Practical Character Physics

This research involved taking animated sequences and adjusting them to be more physically plausible. I think it's a great idea, and in some cases significantly improved the believability of a motion. I also really liked that the artist can decide whether or not to make the adjustments - if a character has a super power or something, he might not always move in physically accurate ways. I think it's really important to give the artist these stylistic choices. Also, I think this was a really great example of how important a character's interactions with its environment are. In the video, it would show an animated sequence, and then show the same sequence with the adjustments made. However, the character was usually just in empty space, and most of us barely noticed a difference. Even though the clip showed the different paths, the motions, at least to my eyes, barely looked different at all. I think a lot of what makes an animation look physically possible or not possible has to do with how objects interact with each other, so it was disappointing for me to see the video and the characters not interact with anything!

3/20 Simulating Balance Recovery

We saw a few videos where characters were walking and had to react and keep their balance after being hit. Some were hit from behind, some from the front; some were merely pushed by an invisible force, and one was even pelted with dodge balls. Some characters did look like they were reacting to being hit, whereas I felt like others looked as if they were trying to avoid being hit, dodging the oncoming forces rather than reacting to them. These motions are pretty similar physically, so what made them look different to me? I don't know. But I find it really interesting how people can differentiate between two extremely similar motions by their intention. When we see a character move, most of the time we are pretty good at guessing what they were trying to do with that motion (so long as the motion is convincing). I don't know what could have been changed in the videos we saw for me to see the correct motions of the characters - reacting vs dodging - but I think it has to do with the fact that, in a simulation, there's really no way to code for a character's intention. Not that I can think of, anyway. You code the simulation to carry out a certain motion, but what if people do that same motion for different reasons? Which reason will be the one audiences see?

3/6 Catching Fly Balls

I found today's topic to be particularly interesting, probably because I'm a huge baseball fan (go Dodgers!). As I mentioned in class, I think the reasoning behind this research is extremely applicable and important, even if actually catching fly balls isn't animated as often as, say, walking. But I think it's really important to realize that searching for the optimal motion is almost always going to be the wrong motion, at least when you're talking about humans. What people do, and what looks natural, is hardly ever optimal. So for this example, people don't immediately move to where the ball is going to land - they adjust. This actually reminded me of a computer game I used to play back in elementary school called Backyard Baseball, so I looked it up to see how the characters caught the ball. Of course, they were pretty clumsy because all of the characters are supposed to be kids, and I think when you're playing the outfield you just click on the character you want to try to catch it. So I thought this game might have a good example of this type of research, but it's so basic (from the YouTube videos at least) that it doesn't seem to have much in common besides baseball. Here's a video, though, in case you want to check it out. I used to play this all the time with my little brother!

Anyway, I was wondering if the research noted any differences between left- and right-handed players, or if they even used test subjects with different dominant hands. When we saw the graph showing where the players walked when catching the balls, it was separated into four quadrants depending on where the ball was landing. The front and back didn't look too different, but I noticed the paths differed from each other more on one side than the other. I thought this might be due to whether the player was left- or right-handed, but it didn't seem like the paper mentioned this at all. It would be interesting to see how a person's dominant hand affects their motions.

3/1 Stochastic Character Animation

I really liked the idea of this paper, being able to create different versions of the same action. I think this is important because we don't all do the same things the exact same way, and if you have one "stock" version of an action, not only is it repetitive but it might not look right on different characters. However, I didn't think it was necessarily successful in creating natural motions. The motions, for the most part, looked physically possible, but sometimes didn't look at all like what someone might naturally do. For example, in the video, there was a character swimming breast stroke, which looked pretty normal. Then we saw the output from this algorithm to reproduce the breast stroke, and it looked so bizarre to me! Yes, the motions were certainly physically plausible, but the timing between the arms and legs looked very strange - not like how someone who really knew how to swim would do. There's an important distinction between what motions are possible and what motions are natural, and this seems to be a problem a lot of people run into. How do you put constraints on your algorithm to only create natural movements? All the time we see outputs that are physically possible, but no one would ever actually do the actions that way. To a computer, they're all the same. I think this is what makes all of this stuff so difficult! There's really no way to separate out the awkward, unnatural motions in your algorithm.

2/28 Video Mocap

This research seems cool, but like an awful lot of work. Obviously it's less work (and looks better) than key framing the whole animation, but there were just so many steps to it! Way more than the other research we've looked at, I think. I definitely don't think it's the best thing for all mocap situations, though it would be useful in situations where regular mocap is impossible. Things that are outside or need large, specific environments, for example. I was thinking maybe swimming. I only have a vague understanding of how regular mocap works, but I don't think it could really work to capture swimming, and obviously someone pretending to swim won't look right. If you had footage of someone swimming, this method might work. Also someone brought up mocapping specific people, such as athletes for their video games. I think this would definitely be a good option for that, since you don't need to bring in the person to physically act for you. Overall I think this method is interesting and useful for some specific situations, but for the most part I think standard mocap is probably better. Of course, I don't know exactly what goes into mocap, such as how you clean the data, etc.

2/23 Human Motion Synthesis with Optimization Based Graphs

I thought the idea of blending between motions or actions was really interesting. If it works well, it could really make things easier for an animator; just tell the character, do this, then this, then this, and not have to worry about what's going on in between. I feel like I say this a lot in these blogs, but this would be really great for video games. Since the character in the game can really do anything at any moment, depending on what the player chooses to do, there has to be transitions between actions. A lot of the time, this is 'easy' since the characters are walking - I feel like the animators can, for the most part, assume the action will come from either walking or standing, something neutral. For a lot of games, this will always be the case. This would be really good, though, for fighting games, so the character doesn't have to come back to neutral between each punch, kick, etc. Unfortunately I'm not sure if I completely understood how this method blends between actions... I got a bit lost in some of the math.

2/16 Perceptional Animation of Body and Hand Motions

Sophie Joerg gave a guest lecture today on her research regarding hand and finger animation. I guess I didn't realize how much finger motions can convey. Hands are obvious - we all talk with our hands, some more than others, because they help get our point across. When you talk with your hands, though, you don't (or at least I don't) think about what your fingers are doing; they just go, whereas you're consciously moving your hands. I think it's because of that that we can so easily notice strange finger motions. We don't think about them because they come so naturally, so when they look wrong it is very obvious.

I was also interested in what Sophie had to say about the idea of the uncanny valley. I've heard of it before, but no one has ever mentioned how it differs from person to person. For example, she showed a clip from The Polar Express and said how strange the characters looked and how bad the critics' reviews were, but honestly I didn't think the characters looked that unnatural. She said that to some people, the characters were creepy but some people didn't mind them at all - I definitely fell into the latter group. I didn't realize how greatly this affects different people.

2/14 Optimal Gait and Form for Animal Locomotion

This class we looked more at using optimization functions to control character motion. I thought it was pretty cool that the functions worked decently for all different sorts of characters: one legged, two legged, four legged with dog proportions, four legged with giraffe proportions, even five legged. Things like this are really useful - one method and one function that works for any sort of character. What I don't like is that it seems like so much trial and error. Randomly changing weights in the objective function until it "looks" right... I don't know of a better way to do it, but I wish there was one. It's like there's so much thought put into the different terms in the objective function, but the weights are completely arbitrary.

2/9 Sampling-based Contact-rich Motion Control

This lecture made me realize something. Derek made a comment about how he thought the character lifted his leg too high when he was sitting in the chair, relaxing, and crossing his legs. I didn't really think it was unnatural though. This made me realize that, so far, we haven't seen anything that takes into account flexibility. How does motion differ between, say, a ballerina, and a regular guy? I'm not saying high kicks or something - that different is obvious - but just basic actions. Like crossing your legs, for example. I'd be willing to bet a dancer would lift her leg higher than most people, because it really doesn't take that much effort when you're flexible. In fact, it feels natural. I think this might be interesting to explore for my project, though I'm not sure how I could create a concrete project out of this. I do think it's interesting that none of the research we have seen has explored this, though.

2/7 SIMBICON

We already talked in class about how awkward characters look when they move with a rigid spine. People simply don't move that way. Another thing I noticed, though, about the simulations is that they all landed on flat feet. This looks silly too, like the characters are constantly stomping around. We've seen other research where they give characters soft feet that bend a little on impact. Of course, something that detailed might not be necessary for this sort of simulation. What I was thinking is that there should be more effort put in to making the character lead with its heel, rather than the entire foot. When we walk, we put our heels down first and the rest of the foot follows, but it seems like the simulations are "aiming" their steps with the middle of the foot, making it land as a whole. Naturally, we walk heels first, but we can lead with our toes instead if we want to. Try walking by landing with your arch first. It doesn't work. If the simulations lead with their heels and allow the rest of the foot to follow in the step, I think the motion can look much more realistic.

2/2 Interactive Character Animation

What I found really interesting in this lecture was the research that used multiple objective functions to create the character animation. For example, the one where the person had to reach for the ball while still keeping himself balanced. I think this technique can be really useful. Yes, you want the character to perform a certain action, but you also want him to be physically plausible, and this is a good way to do it. Weighting physical constraints higher in your function than the desired action is an interesting way to maintain a character's believability. While we might not realize this, I think this is pretty similar to the way people think when we do certain actions. For example, imagine you're standing on a ledge and you want to get down. Reaching the ground is the action we want to perform, but that's not the only thing in our internal objective function. We also take into consideration our physical capabilities as well as our own safety. If the ledge is only a couple feet off the ground, we might just jump, but if the ledge is a lot higher, we think of another way to get down safely. We might sit down on the ledge and then jump, or look along the sides for some stairs. Sure, we don't think about complicated physics when we're deciding how to act, but we do take many things into consideration.

1/31 Automated Character Locomotion

I thought this paper was really impressive. Being able to automatically change a character's walk to suit the terrain can save a ton of time for animators; instead of animating a different walk for each situation, you only have to do one! However, I'm pretty skeptical of using the data from one character for another, ex. using data from an adult human for a humanoid animal or a small child. I suppose maybe it could work for an animal doing impossible things, since we don't really know what those actions should look like. Going with the example we talked about in class, no one has seen a real alligator dance, so I think audiences will be forgiving. A dancing alligator doesn't have to look completely realistic because it's not realistic, so maybe this type of technology could work using adult human data. I'm far more skeptical about using this to animate a baby or a toddler. Yes, they're all human and have the same joints and physical capabilities as adult humans, but they simply don't walk the same way. We all know what toddlers look like when they run. They waddle because they still haven't quite gotten a hang of what they're doing, not to mention they're wearing a diaper. An adult might be able to imitate this, but I think it would look really weird if you tried to use the data from an adult walking regularly to animate a toddler. Especially since everyone knows what toddlers look like, it would look extremely unrealistic, and honestly probably a bit creepy.

1/26 Direct Control Interfaces

Motion capture can definitely help to animate characters more quickly, but in lecture today I realized how many limitations there are to this technology. A lot of things are physically impossible for the actor to do without creating a whole set for him/her to perform in. For example, I think someone in class mentioned pushing a car. I have far too much experience pushing cars, and I can tell you that you can't convincingly mime that action. When you push a car, you're leaned over really far, throwing your whole body into it - you simply can't do this without something to lean on! Another thing, which we talked about a lot, is showing force. I'm saying "leaning" onto the car, but really that's the wrong word - you're pushing it with your whole body. To a motion capture system, leaning and pushing look pretty similar. Your body is in almost the exact same position, so why do the two look different? People can differentiate easily between the two actions. Is there a way to make the motion capture system automatically know when a person is exerting a lot of force, or just leaning casually? It's a pretty difficult question. I think what we saw today, though, would be really cool for games. The video mentioned actors interacting on screen even though they were performing miles away from each other, and I immediately thought it would make for cool multiplayer games. Imagine playing tennis online with your friend across the country!

1/24 Directing Physically Based Interactions

We saw some examples today of a user moving the mouse to direct a character's head (for the lamp) or center of mass (for the starfish) in order to make it move. I'm sure some artists might not like this because it doesn't allow complete control of the character. The directions given from the mouse are always secondary to physical constraints, which can help the character to look more believable, but sometimes, especially with "cartoony" characters, physical constraints aren't what we want. However, I think for the average user this system is intuitive. Especially with the lamp, it reminded me of something you might see on an iPhone game or something - moving the character through an obstacle course by directing it with your finger on the screen. For most people, I think this system would be pretty easy to use. Also, people tend to be pretty forgiving of the graphics in games. Things don't look perfectly realistic in games, and people know this. For movies, though, the standards are definitely higher. In a game, we have to pay attention to strategy, whereas in a movie our attention isn't deviated away from the visuals as much. I think that's why some artists might be critical of the center-of-mass method of directing the characters. Artists want complete control of the look of their characters, so this system is a good starting point for animation but I don't think it can work at this point as the entire means of animating a character.

1/19 Noise, Variability In Movement

I think it's very important that characters are given some additional noise. When they're moving, noise might not make a huge difference, but I'm sure we all can agree that a character standing completely still looks unnatural. When real people stand "still," they shift their weight, touch their clothes, and look around. It's impossible to stand completely still, which is why adding a little bit of noise to an animated character can make all the difference. What I did notice, though, is that there's definitely such a thing as too much noise. We looked at some of Ken Perlin's interactive characters, such as the Emotive Virtual Actors, the red/green couple, and their level of noise works. Granted, they're pretty simplistic characters, but I think the way they move when left alone looks believable. Then if you look at the Face Demo, the close up of a woman's face, it's like she's moving nonstop! If you tell someone to look straight ahead, sure, they will move a little, but no one naturally moves as much as this woman does when told to hold still. I learned from these examples that while noise can help the character not to look robotic, it's easy to overdo it and stop being natural and believable.