Chapter 20. Selecting the Target

Key Topics

• Case Study

• Rationale

• Module Design

• Implementation

• Application

• Evaluation

This chapter applies multilayer perceptrons (MLPs) to a difficult problem. Animats are going to learn to estimate the damage of a rocket on the enemy. This capability enables them to pick targets that are most likely to injure the opponent—either directly or via splash damage. Better target selection also prevents the animats from blowing themselves up.

The next few pages cover all aspects of the problem, from an informal description to the implementation and evaluation.

This chapter covers the following topics:

• A case study of the target-selection problem, explaining what properties of the situation affect the decision of where to shoot

• A few ideas for the development, notably to tackle the experimentations if things go wrong

• The interfaces to the multilayer perceptron, both for runtime and initialization

• The implementation and how the data structures and algorithms come together

• How the application phase tackles the problem, what issues arise, and how the perceptron fits into it

• The resulting animat and the target-selection behaviors

At the end of this chapter, the animats will be capable of handling rocket launchers as lethal weapons—without hurting themselves too much.