Eva Schiffer

Group Motion Graphs. Yu-Chi Lai, Stephen Chenney, and ShaoHua Fan.

Summary:

This paper discusses the construction and use of motion graphs to describe complex flock motions. Because the motion clips of the flock are calculated in advance, this method is computationally efficient at runtime.

Problem:

How does one generate plausible motion for a large flock in a realtime application?
How can one constrain the motion of a large flock at runtime with the least amount of processor time spent on this task?

Method:

This paper essentially builds motion graphs for flocks that are similar to those used for motion capture data. The main difference is that the individual clips in the flock motion graph are generated computationally rather than being recorded. Also, the flock motion graph makes the assumption that individual agents all behave with the same set of rules, so when transitioning from one flock motion graph to another, it does not mater which agents from clip one map to which agents from clip two, so long as the difference between the clips is not great.

Key Ideas:

Pre-calculation can generate dense flock motion graphs that allow for fast flock constraint satisfaction at runtime.
Using pre-calculated flock motion graphs can allow for much larger simulations (in terms of agents) since the individual interactions of the agents have already been determined.

Contributions:

The system for making and using group motion graphs presented in this paper is a powerful way to form complex agent simulations with little calculation overhead. However the motion graphs do take considerable amounts of space in memory, so it would be important to assure that the system using them has such memory to spare.

Questions:

I had no significant questions about this paper.