Chapter 18. Dealing with Aiming Errors

Key Topics

• Momentum and Friction

• Dealing with Errors

The preceding chapter covered a specific kind of neural network known as perceptrons. These are capable of solving simple problems using linear approximations. To help demonstrate the theory, the next few pages apply perceptrons to improve shooting capabilities. Many problems could be adapted for this purpose, but this chapter focuses on creating realistic yet efficient aiming, which involves problems that are linearly approximable by design.

The sections in this chapter cover one problem each, dealing with realism and then effectiveness:

• One perception improves the aiming realism by smoothing the actions. The friction and momentum (in the mouse) is modeled. A perceptron provides a fast linear approximation of these aiming errors for the animat.

• To deal with these imprecisions, another perceptron solves the inverse problem by trial and error. The artificial intelligence (AI) collects information online and retrains the perceptron accordingly.

At the end of this chapter, the animats will make plausible aiming errors, but also correct the turning angles to take them into account. This provides an ideal compromise between realism and effectiveness at shooting (with adjustable skill settings).