I am Garrett Udstrand (udstr013@umn.edu). This is my website submission for Project 2 in CSCI 5611. The example video of the project can be viewed below.
My submission, this time around, is pretty bare bones. I simply simulate a cloth in a 3D environment and give the user a camera to navigate the environment. It runs pretty fast, only dipping to 45 fps at lowest on my laptop, which doesn't have the greatest of GPUs.
Running the project will start you off with a cloth in front of you and an obstacle underneath it. Hitting the spacebar allows updates to happen and makes the cloth fall. Hitting 'r' resets the simulation. The WASD keys are used to move the camera around, and you can rotate the camera by using the arrow keys. Dragging the mouse around while paused allows you to move the obstacle, so you can try out the simulation against many different obstacle positions.
This one was required, so, of course, I implemented it. As you will see in the next section, I extended this multiple rope simulation to be a full-fledged cloth simulation, so there are no literal ropes being displayed in the simulation, just rope calculations being used implicitly in the cloth simulation.
I added horizontal ropes as the instructions directed and was able to create a simulated version of a cloth. I use a fairly standard example, with the cloth falling onto a ball and making satisfying cloth-like motions due to hitting the ball. I placed the ball slightly off center, so you get this really interesting side-to-side movement in the cloth that reminds me of clothes flowing in the breeze. This cloth simulation is seen throughout the video.
The simulation is done in 3D and a user can use the camera to move around it. You simply hit the WASD keys to move, and the arrow keys to change the camera's rotation. It's almost like a video game, but using arrow keys rather than the mouse. You can see the camera being used to get different angles of the 3D environment at approximately 10 seconds in and 25 seconds in. You can see the camera being used more dynamically below, with it being moved around while the simulation is going, just to see what the cloth looks like from different angles.
By dragging the mouse around, you can move the obstacle to see the cloth interact with it in different ways. I only allow you to move it when the simulation is paused, however, as moving the ball while the cloth is moving can lead to glitchy behavior. I should also note that if you move the ball so that the cloth and the ball are inside one another, the simulation will also glitch out and break. You can see the ball moved at approximately 19 seconds into the video and 36 seconds into the video. You can see an example of me breaking the simulation below.
I tried to match up my cloth's speed to something like a big, billowy blanket. I wanted the cloth to feel weighty and comfy when seen in movement, which I think I certainly achieved, but that's subjective, I guess.
The simulation was written using processing, and the camera is mostly copied from the camera.pde example provided on Canvas. The in-class rope activity was used as a base for creating the simulation, but it has been significantly modified. Almost none of the original code remains.
The source code can be found on GitHub here. You can also download a ZIP of the repository here.
While we have discussed collisions quite a bit in class, it's a bit more difficult to get good looking collisions with the cloth. It deforms and kind of stretches and this can be difficult to work with. If the cloth deforms too much on the sphere, and the push is not strong enough, the sphere will show through the cloth and it will look bad. However, setting too strong of forces can make the cloth float. It can also make it jitter, which is a problem still kind of present in my final submission.
A solution I was interested in pursuing was testing the intersection between triangles and the sphere rather than between individual points. Because there are configurations where the individual points of a triangle will be outside the sphere, but the actual rendered triangle is still clipping it. To test this intersection, you'd simply treat the triangle as a plane and use the distance from a point to a plane equation. If the distance calculated is less than the radius of the sphere plus some tuning value, you push the plane outwards until that distance is greater than those two values. The problem with that solution, though, is it's really hard to implement pushing the plane outwards. How exactly you change the points, so that they remain relationally the same, but still move the distance of the plane from the center of the obstacle out, is not exactly clear, and my linear algebra knowledge was not good enough to make it happen in time.
As you probably saw in the features section, the collision system sort of breaks if the obstacle being collided does not start on the outside of the sphere. I spent forever debugging this, and I couldn't come to a particularly satisfying conclusion on it.
Overall, it mainly came down to using the same basic logic, but tinkering quite a bit with the parameters until the collision looked good and produced a good looking cloth movement.
If you don't have enough polygons to render the cloth, it can look very angular and it breaks the immersion quite a bit. The problem is that adding too many polygons can give your computer quite the hit, and will reduce how smooth you can make the overall motion.
Just like before, there was no supremely elegant solution to this problem, and, if there is one, I simply didn't have the time to work it out. Instead, I just adjusted the parameters until I had a good balance between the amount of polygons and the smoothness of motion.
I made attempts to include both drag and more proper lighting in this simulation, but I had to scrap both because they were glitching out on me, which I believe was due to the cross product. For the drag and for the lighting, we want our normal in a particular direction, so that the calculations work properly. However, the cross product only chooses its direction via the right-hand rule, and I had situations where the cross product would suddenly flip, and cause either the lighting or the drag to wig out and look bad. Perhaps, with more debugging, I could've fixed it, but I decided to pursue other avenues to get a nice simulation.