Last Modified 23 March 2008

Behavior Oriented Design (BOD)

Behavior Oriented Design (BOD) is a methodology for developing control of complex intelligent agents, such as virtual reality characters, humanoid robots or intelligent environments. It combines the advantages of Behavior-Based AI and Object Oriented Design.

Developing a system under BOD involves creating two types of program structures:

A library of Behavior Modules. These are a set of conventional OOD objects representing an interdependent set of parallel perception/action/learning modules. The modules generate and control the actions of the agent, process its perception and store and maintain any memory required if the agent learns. Only one library is needed for platform, agents can share instances of this code.
A POSH dynamic plan for action selection. This is a flexible, hierarchical data structure which provides structured intelligent control by setting an individual agent's priorities.

BOD agents can be looked at from two perspectives. You can think of the behavior modules as providing the primitives for the POSH plans (the plan primitives are based on methods from the behavior objects), or you could think of the plans providing behavior arbitration (a form of action selection) when the actions recommended by the behavior modules would conflict.

Above all, BOD is a methodology. Its purpose is to make developing agents as easy as possible. This requires keeping the agents as simple as possible, so that they are easy to understand, develop and maintain. The BOD methodology consists of

an initial decomposition process to help get you off to the right start,
a set of heuristics for determining which representation is best for representing an aspect of agent intelligence clearly, and
an iterative development methodology with rules for refactoring the system to keep it as simple as possible while scaling its capabilities.

Software / code using BOD and providing POSH for python, jython and lisp is available for free download from the AmonI Software Page. This includes two major sets of behaviour libraries for the python/jython POSH:

BOD / MASON for agent based modelling (social simulation), and
BOD / UT for building characters in the first-person shooter computer game, Unreal Tournament.

We also are working on a BOD IDE, ABODE.

Note we are always happy to have more BOD Developers (that page also takes bug reports.)

For more details about using BOD, see my papers on Designing Intelligent Systems.
For our current research on BOD, see the AmonI Projects Page.
The AIGameDev.com review of BOD.
For other people's AI architectures, see my related research page.
Code created under BOD (including a demo / tutorial) is available from the AmonI Software Page.
BOD-related research at other universities:

Cyril Brom supervises a group that uses POSH extensively. Their Pogamut includes a java version of POSH (and a netbeans-based lap-file editor) for computer games (UT2004) at the Charles University in Prague (2005-present).
Mark Humphrys' group at University College Dublin implemented a version for their World-Wide Mind project, though it isn't currently online (2004).
Torbjørn Semb Dahl has used a version of BOD in an ALife competition (2001).
Send me a URL to get added to this list.

page author: Joanna Bryson