---

Общие сведения

Архитектура Subsumption разрабатывалась в 1995-1991 гг. для когнитивной робототехники Родни Бруксом  (Rodney A. Brooks) в Лаборатории искусственного интеллекта (CSAIL)  Массачусетского технологического института, Кембридж, Массачусетс, США. Это – реактивная, многослойная с приоритетной селекцией реакций без представления знаний.

Subsumption Architecture

In the mid-1980's Rodney Brooks touched off a firestorm of interest in autonomous robots with the introduction of the Subsumption architecture 1 [Brooks86].

Brooks has developed a methodology and architecture — the subsumption architecture — for controlling a number of mobile robots known as artificial beings [Brooks86] or Creatures  [Brooks91].

The subsumption architecture is a layered controlled system in which each layer, rather than implementing some individuated control function such as perception, memory, attention, or judgement, realises instead a particular task-achieving behaviour or domain-specific competence (for example, obstacle avoidance, random wandering).
This way of dividing up an agent’s faculties is often referred to as a vertical decomposition, in contrast with the more conventional horizontal method of faculty decomposition, used by, among other things, traditional planning systems.

Each layer in the subsumption architecture is composed of a fixed-topology network of FSMs, together with one or more data registers and internal timing units.
Layers communicate with each other using fixed-length messages over low-bandwidth channels or wires. The FSMs within a layer can be made to change state upon the arrival of messages from other layers or after the expiration of designated time periods.
Layers operate asynchronously and in parallel with each other and do not employ any shared global memory.
Control is layered (non-hierarchically) with higher level layers subsuming the roles of the lower level layers when they wish to take control.
In particular, layers are able, for finite pre-programmed time periods, to substitute (suppress) the inputs to and remove (inhibit) the outputs from lower level layers, and thus affect the normal flow of data within the layers.

Careful programming of this inter-layer control will have the overall effect of “biasing” the agent’s actions toward achieving its higher level goals while still attending to its lower level or more critical goals.

Fig. 1. Abstract representation of subsumption architecture,
with the higher level layers subsuming the roles of lower level layers when the sensory information determines it. [Brooks99]

---

Robots controlled using the subsumption approach have been successfully deployed for such tasks as indoor room exploration, map building, and simple route planning [Brooks91].

The Cog Project: Building a Humanoid Robot

The motivation behind creating Cog is the hypothesis that:
Humanoid intelligence requires humanoid interactions with the world.
Cog Project Publications

Subsumption for the SR04 and jBot Robots, DPRG website

---