SBR Overview

TinyCog is an implementation of Scene Based Reasoning (SBR), which is a hybrid cognitive architecture based on spatial-temporal planning using Hierarchical Task Network (HTN). On this base we propose a deduction system for Description Logics (DL) and mechanisms to implement plan, social, abstract and introspective reasoning.
SBR rests on the concept of "scene" as the main data-structure across a wide range of subsystems. Scenes represent real-world 3D scenes as well as "mental images". On the lowest level, scenes are implemented as 3D scene graphs (as in computer gaming), while behaving on higher levels more like Minsky frames, representing objects and their relationships. Scenes are the key to meta-reasoning on plans by representing plans themselves as scenes.
The spatial-temporal structure of SBR, while being symbolic to some degree, allows for the application of statistical and connectionist formalisms with the well-known benefits in terms of training friendliness, performance and scalability.

This figure shows a schematic overview of the SBR layer cake: