gale.ai.scripting module

This file contains a small data-driven “scripting” layer for AI: a Registry of named actions/conditions/tests/weights, plus build_behavior_tree and build_decision_tree, which recursively turn a plain, JSON-friendly dict into a gale.ai.behavior_tree.BehaviorTree or gale.ai.decision_tree.DecisionTree. This lets a designer describe an agent’s decision-making as data (hand-written, generated, or loaded from a JSON file) instead of writing Python subclasses for every leaf, without needing a custom text-based language and its own parser.

Author: Alejandro Mujica (aledrums@gmail.com)

class gale.ai.scripting.Registry

Bases: object

Holds the named building blocks a behavior tree/decision tree spec can refer to by name instead of by direct Python reference: actions (agent, dt) -> Status/Any, conditions/tests (agent) -> bool, and named weights for RandomDecisionNode branches.

Usage example:

registry = Registry() registry.register_action(“patrol”, lambda agent, dt: agent.patrol(dt)) registry.register_condition(“enemy_visible”, lambda agent: agent.can_see_enemy())

register_action(name, function)

Parameters:

• name (str) – Name a spec can refer to this action by.

• function (Callable[[Any, float], Any]) – Callable(agent, dt) implementing it.

Return type:

None

register_condition(name, predicate)

Parameters:

• name (str) – Name a spec can refer to this condition by.

• predicate (Callable[[Any], bool]) – Callable(agent) -> bool implementing it.

Return type:

None

get_action(name)

Parameters:

name (str) – The name an action was registered with.

Returns:

The registered callable.

Raises:

KeyError – If no action was registered under name.

Return type:

Callable[[Any, float], Any]

get_condition(name)

Parameters:

name (str) – The name a condition was registered with.

Returns:

The registered callable.

Raises:

KeyError – If no condition was registered under name.

Return type:

Callable[[Any], bool]

gale.ai.scripting.build_behavior_tree(spec, registry)

Recursively build a behavior tree node from a spec such as:

{

“type”: “selector”, “children”: [

{“type”: “sequence”, “children”: [

{“type”: “condition”, “name”: “enemy_visible”}, {“type”: “action”, “name”: “attack”},

]}, {“type”: “action”, “name”: “patrol”},

],

}

Recognized “type” values: “sequence”/”selector” (need “children”: a list of specs), “parallel” (also accepts “success_threshold”/ “failure_threshold”), “inverter”/”succeeder”/”failer”/ “until_success”/”until_failure” (need “child”: a single spec), “repeater” (needs “child”, accepts “times”), “cooldown” (needs “child” and “duration”), “action” (needs “name”, looked up in registry), “condition” (needs “name”, looked up in registry).

Parameters:

• spec (Dict[str, Any]) – The spec to build a node from.

• registry (Registry) – Where “action”/”condition” names are looked up.

Returns:

The built Node. Wrap the result in gale.ai.behavior_tree.BehaviorTree to tick it.

Raises:

KeyError – If spec’s “type” is not one of the recognized values above, or an “action”/”condition” name is not registered.

Return type:

Node

gale.ai.scripting.build_decision_tree(spec, registry)

Recursively build a decision tree node from a spec such as:

{

“type”: “decision”, “test”: “low_health”, “true”: {“type”: “action”, “name”: “flee”}, “false”: {“type”: “action”, “name”: “attack”},

}

Recognized “type” values: “decision” (needs “test”, “true”, “false”), “random” (needs “branches”: a list of [spec, weight] pairs), “action” (needs “name”, looked up in registry).

Parameters:

• spec (Dict[str, Any]) – The spec to build a node from.

• registry (Registry) – Where “test”/”action” names are looked up.

Returns:

The built DecisionTreeNode. Wrap the result in gale.ai.decision_tree.DecisionTree to evaluate it.

Raises:

KeyError – If spec’s “type” is not one of the recognized values above, or a “test”/”action” name is not registered.

Return type:

DecisionTreeNode