Planning/Reactivity Dichotomy

While planning and reactivity are not mutually exclusive characteristics of an architecture, in the sense that a given architecture may exhibit both planning and reactivity, they are characteristics that serve complementary purposes and typically are implemented in distinctly different fashion. Many agents do have implementations of both planning and reactivity, but doing so requires additional abilities, since reactive actions often change the enviroment and may render invalid assumptions necessary for the execution of a plan. To compensate, a reactive planning agent should be able to replan based on unexpected changes in the environment, and it must be able to recover from interrupts. Because of this difficulty, some architectures ignore either planning or reactivity and focus entirely on one of these.

The two characteristics together are responsible for determining an agent's future course of action; a sequence of an agent's actions is usually composed of elements of a plan, punctuated by reactions to unforeseen situations. Planning is concerned with positive actions -- actions that are directed toward achieving a particular goal; while reactivity has a more negative focus: NOT running into the wall, NOT getting eaten by the tiger, NOT getting damaged and/or destroyed. Since planning requires a certain amount of thought (dare I say 'creativity'?) on the agent's part, and, in most cases, there are minimal time constraints on the planning process, planning is cognition-intensive and, therefore, relatively slow. In contrast, reactivity demands fast processing and therefore a minimum of cognitive commitment.

Planning and reactivity may also be compared to Newell's automatic and controlled behaviors. Reactive agents must exhibit some automatic behavior, while planning is responsible for the agent's controlled behavior, which requires some forethought in order to accomplish.

Back to List of Dichotomies