Skip to content

README-appendix.md

Latest commit

 

History

History
76 lines (62 loc) · 3.14 KB

README-appendix.md

File metadata and controls

76 lines (62 loc) · 3.14 KB

Embabel Agent Goal Planning System

Key Components and Their Relationships

Action

  • Core executable unit that performs a specific task
  • Implements GoapAction interface and the ActionRunner interface
  • Has preconditions (what must be true before it can run)
  • Has effects (what becomes true after it runs)
  • Contains inputs and outputs defined as IoBinding objects
  • Can consume tools via the ToolConsumer interface
  • Has a cost (computational expense) and value (benefit of execution)
  • References transitions that define how action affects the system state

Condition

  • Represents a named predicate that can be evaluated
  • Can be TRUE, FALSE, or UNKNOWN (via ConditionDetermination)
  • Has a cost associated with evaluation
  • Can be combined using logical operators (AND, OR, NOT)
  • Used to define preconditions for Actions and Goals
  • Evaluated against the current world state via ProcessContext

Blackboard

  • Shared memory system for maintaining context
  • Implements Bindable interface for variable binding
  • Stores objects in an ordered list accessible by name or type
  • Can spawn independent child blackboards
  • Can track condition values explicitly
  • Acts as the repository for all data and objects in the system
  • Used by the World State Determiner to derive current conditions

WorldState

  • Represents current state of the system as a map of conditions
  • Key component in GOAP (Goal-Oriented Action Planning)
  • Used to determine which actions are achievable
  • Can generate variants with different condition values
  • Foundation for planning algorithms

Plan

  • Sequence of Actions to achieve a Goal
  • Has associated cost and value metrics
  • Created by Planners (like AStarGoapPlanner)
  • Represents the path from current state to goal state

GoapPlanningSystem

  • Collection of available Actions and Goals
  • Tracks known preconditions and effects
  • Used by planners to construct valid plans

How They Work Together

1. Planning Process

  • The WorldStateDeterminer reads the Blackboard to determine current conditions
  • The AStarGoapPlanner uses A* algorithm to find optimal sequence of Actions
  • Each Action's preconditions are checked against WorldState
  • The planner constructs a Plan with lowest cost path to satisfy Goal conditions

2. Execution Flow

  • Actions read inputs from and write outputs to the Blackboard
  • Actions can only execute when their preconditions are satisfied
  • When executed, Actions update the WorldState through their effects
  • The system re-evaluates Conditions after each Action

3. State Management

  • Blackboard maintains the actual objects and data
  • WorldState maintains boolean conditions derived from Blackboard
  • Conditions provide the evaluation logic to determine truth values
  • Plans are re-assessed after each Action executes

This architecture implements Goal-Oriented Action Planning (GOAP), allowing the system to dynamically determine sequences of actions to achieve goals based on current conditions.

Visualization

To review key components visually: