|Classification of Complex Robot
MAS622J Final Project - Aisha Walcott - Fall 2006
Methods and Results
To determine our set of features, we first note that each data file is an instance of the Swarm. As the Swarm executes each behavior, they evenutally reach the intended goal state (eg. Bubble Sort) which is the steady state. There is a unique steady state for each behavior. Features are used to determine if and when the Swarm reaches a steady state. Each steady state/high level behvaior is defined by its own set of features. The figures below illustrate some of the features used for each behavior.
Figure 1: Example Swarm Behavior Features
The Feature Vector
Fourteen features were defined from the data resulting in a 16-D feature vector. The features were primarily derived from the convex hull density feature, the line fit feature, the distance to source feature, and the dot product to source feature. These were calculated for each swarm instance at steady state. In addition, statistics such as the average and the variance were used in the feature vector.
x = [avg_density,