Jump to the Architectural Overview Menu page.
Jump to the Home page.

An Overview of the Brooksian Architecture

In the quest for intelligence, AI has focused on issues such as knowledge representation, planning, reasoning, and learning. Rodney Brooks has a divergent philosophy for the constructing intelligent agents. In looking at lower animals, one sees that most of their activity is concerned with rather mundane aspects of merely existing in the world. For example:

there seems to be a very close connection of sensors to activity;
there are pre-wired patterns of behavior;
very simple navigation techniques are used;
they are almost characterizable as a deterministic machine;

Very little of this activity can be mapped onto what is currently being developed in AI. This has led him to construct simple, behavior-driven agents, based on these properties, that can work robustly in complex, dynamic environments. Traditional AI is based on duplication human functionality. Brooks' view is orthogonal is focused on duplicating behavior. Once this is achieved, they can be enhances by incrementally adding more intelligent behaviors.

The characteristics of these architectures are guided by a few simple assumptions:

there is no need for representation;
the world is its best model;
nothing in the architecture is centralized;

Characteristics

Sources and References

Brooks, R.A., "How to Build Complete Creatures Rather than Isolated Cognitive Simulators" in Kurt VanLehn (Ed.), Architectures For Intelligence, pp. 225-39, Lawrence Erlbaum Associates, Hillsdale, NJ, 1991.

Brooks, R.A., "Integrating systems on behaviors," SIGART Bulletin 2, 1991, pp. 46-50.

Brooks, R.A., "Intelligence Without Representation," Artificial Intelligence, vol. 47, pp. 139-59.

Brooks, R.A., " Intelligence Without Reason", available by ftp from publications.ai.mit.edu/ai-publications/1991/ as AIM-1293.ps.Z.