SOAR: Cognitive Architecture for AI Problem-Solving and Learning
To lay some groundwork for building cognitive architectures
Assume we want to build a simple cognitive robotic agent tasked with navigating a maze to reach a goal. One possible way to achieve this is the application of a cognitive architecture called SOAR [1]. SOAR has been in research since 1983 and was developed as a theory and computational model of human cognition.
The SOAR architecture operates on the Problem Space Hypothesis[2]. The Problem Space Hypothesis is a psychological theory that suggests that problem-solving involves the creation and manipulation of mental representations (some call it a world model), or problem spaces, in the mind.
These problem spaces are commonly a combination of Goals (G), Problem Spaces (P or PS), States (S), and Operators (O or Op). SOAR suggests that all goal-directed behavior can be viewed as a search through a space of possible problem states to achieve said goal.
Now to our short example of designing a simple robotic agent that can navigate a maze.
Goal
The goal is to complete the maze.
State
In SOAR, a "state" represents the current situation or condition of the system or environment being modeled. It encompasses all the relevant information that describes the system at a particular moment. This information about the environment is typically organized into a hierarchical structure to explain dependencies. States can include facts about the environment, and the agent's current goals, beliefs, and perceptions.
In our quick example, it should include
Current position of the robot in the maze.
Location of obstacles in the maze.
Goal position that the robot is trying to reach.
Here is a pseudo-code of how this might look like: