# Some Ideas for Project 1 (and later)

Unfortunately, we need to pick project 1 topics before we've seen too
much in class. So here are some ideas.

These ideas fall into 3 categories. All of these project types involve
some reading, some implementing, and some experimentation. The balance
just shifts:

1. Mini-research projects

   These projects ask you to look at something that is not well
   covered in the literature. Over the course of an initial project,
   you should be able try some experiments and think things
   through. If the project continues, it _might_ lead to some new
   result. 

2. Practical Implemetation projects

   These projects focus on implementing a solution to a problem that
   has a known solution. Usually, this will be some problem of
   practical importance. These projects _may_ lead to research results
   if you come up with some better way of solving an important problem
   (these are the best kind of research results!) or provide
   infrastructure for other things.

3. Survey Projects

   The idea of these projects is to gain an understanding of what is
   out there for a certain problem. These projects will require you to
   read the literature, and to follow the chain from the basic ideas
   to the current practice. The implementation component of these
   projects will be to make sure you understand some of the basic
   methods, and to see a road to something more complex.

   If you choose a project of this form, you will likely be tapped as
   the "expert" on the subject and be responsible for leading class
   discussion. (everyone will need to lead a class discussion at some
   point in the semester)

These categories are fuzzy - any project's focus can be altered to fit
it into one more than another. However, thinking in terms of these
style categories can make it easier to pick something.

# Mini-Research Ideas

- Gaze Direction Control for motion capture data. Can we figure out
  where the person was looking? Can we manipulate it?

- Blending "Dynamics": most work on blending considers that the
  weights don't change (except for transitions, where the weights
  change fast). Can we figure out how changes in the blend weights
  lead to motions and use this to figure out if/how we can alter the
  blend weights?

- Flexible motion graphs: motion graphs currently pick single points
  of connection and exact placement of the clips. Coming up with new
  methods to represent ranges of legal connections can lead to more
  responsive motion graphs that can be controlled more precisely.

- Better transition finding / generation: right now, the methods that
  are used for determining when transitions can be made in a motion
  graph (or whether two motions can be blended) are quite simple so
  automatic thresholds are elusive. 

  Also, as the method for generating the blends/transitions become
  more sophisticated (say, including some cleanup or physics), the set
  of methods used to determine whether or not a transition might be
  good should evolve.

- Comparisons of blended and real motions. Extensive tests,
  particularly of speed changes in running motions, may be possible
  using data we may be able to get through the Physical Therapy
  department. 

- Stylized Rendering of cloth and hair: is it possible to create
  drawings or artistic representations of things without having to do
  detailed simulation?

# Practical Implementation Projects

- Implement Snap-Together Motion (for authoring run-time characters)

- Implement a fast browser for motion libraries (including methods for
  thumbnailing motions)

- Implement conversion tools to adapt motions between different
  skeletons (this is a suprisingly hard thing to do well!). A nice
  "research" problem is to figure out how to simplify things in a good
  way based on how the motion will be viewed.

- Motion Re-Sampling: given a set of blendable motions, produce a new
  set that is more "uniform" (for example, make turns at uniform
  increments). 

- "Automatic" fitting of skins to skeletons. Automating this is in
  some sense impossible (as it requires creative choices), but there
  are many "research" things that have not been explored much yet,
  such as taking the motion into account when designing the skin to
  avoid artifacts (such as self-intersection).

- Pipeline development for motion capture. At least two groups on
  campus (in Physical therapy and Kinesiology) have motion capture
  equipment. It may be possible to get access to these, although we
  would need to develop a pipeline for converting the data into a
  useful form.

# "Survey" Project Topics

- Cloth Simulation

  Implement the core methods to demonstrate their differences, and
  understand how more evolved methods have progressed.

- Smoke / Fluid Simulation

  Begin with the foundational papers, and follow through towards the
  more "modern" methods that use sophisticated numerical methods that
  are probably beyond the scope of a class project. Implement the
  basic methods and demonstrating their uses and limitations.
  
- Deformable Object Simulation

  Read through the literature to see how the new "real-time" methods
  work. Compare recent methods 
  Implement one (or more) to demonstrate what they are capable of.