Need-based AI in JavaScript

From RogueBasin
Revision as of 22:40, 24 September 2013 by Ondras (Talk | contribs)

Jump to: navigation, search

Warning: this is a work in progress! Do not read until finished.

This article describes an implementation of a need-based AI system in JavaScript. It is somewhat similar to Need driven AI.


We are going to build the AI logic based on the famous Hierarchy of needs. To keep the code clean, all stuff related to AI will be put into a separate (pluggable) JavaScript object "AI". Mixing this AI with a Being will be shown later.

First of all, let's define some "needs" our beings shall consider:

this._needs = {
    survival: 1,
    health: 1,
    satiation: 1,
    revenge: 1

Survival is the basic need to survive, i.e. to maintain at least a minimal amount of hitpoints. Health is a need to be healthy, to regenerate as many hitpoints as possible. Satiation represents the need to fight hunger. Finally, Revenge is a need to avenge any damage that was done to us.

These needs are initially set to 1, meaning "satisfied". This article will only use 1 and 0 values (0 meaning unsatisfied), but for more complex scenarios, float values can be used to represent partially satisfied/unsatisfied needs.

Observing need change

The AI needs to do two major things: watch how the values influencing individual needs change and act accordingly. We can leverage the language used (JavaScript) to perform some major monkey-patching (code surgery). Assuming that our Being object exposes the canonical damage and heal methods, the AI part can implement its own version of these functions:

Being.prototype.heal = function() {
    this._hp = 10;
AI.prototype.heal = function() {
    this._needs.survival = 1; = 1;
Being.prototype.damage = function(attacker) {
AI.prototype.damage = function(attacker) {
    this._needs.revenge = 0; = 0;
    this._needs.survival = (hitpoints < threshold);

With this in place, let's implant the AI code directly onto the being, creating a wrapper function that calls both the original and the AIfied method:

var AI = function(being) {
    this._being = being;
    this._being._ai = this;
AI.prototype._hook = function(func) {
    var original = this._being[func];
    var ai = this;
    this._being[func] = function() {
        var result = original.apply(this, arguments);
        ai[func].apply(ai, arguments);
        return result;

This approach is especially useful as it allows the original Being to exist without any knowledge of an AI. When an AI is created it modifies the underlying being by adjusting all necessary methods.


Our Being has a public act method, which gets called every time the being takes a turn. This method does not do much: it adjusts periodic counters (such as the hunger value), but does not actually perform any actions - the default Being is dumb. Let's hook into this method as well:

var AI = function(being) {
    /* ... */
AI.prototype.act = function() {
    /* observe changed values */
    if (this._being._hunger > 2) { this._needs.satiation = 0; }
    /* act! (we are paid for this...) */

How does the decision process look like? We need a prioritized list of needs, to know which of them must be satisfied before another. This list is used to find unsatisifed needs and to try satisfying them in a correct order:

var priorities = ["survival", "satiation", "revenge", "health"];
for (var i=0;i<priorities.length;i++) {
    var need = priorities[i];
    if (this._needs[need]) { continue; } /* already satisfied */
    if (this._satisfy(need)) { return; } /* managed to satisfy */
AI.prototype._satisfy = function(need) {
    switch (need) {
        case "hunger":
            if (!haveFood) { return false; /* cannot satisfy */ }
            return true;
        case "survival":
            if (haveHealing) {
            } else {
                /* move away from the danger... */
            return true;
        /* ... */
Personal tools