The goal of action-based planning is to determine how to decompose a high level action into a network of subactions that perform the requisite task. Therefore the major task within such a planning system is to manage the constraints that apply to the interrelationships (e.g., ordering constraints) between actions. In fact, action-based planning is best viewed as a constraint satisfaction problem.
The search for a plan cycles through the following steps: choose a constraint and apply the constraint check; if the constraint is not satisfied, choose a bug from the set of constraint bugs; choose and apply a fix, yielding a new plan and possibly a new set of constraints to check.
In contrast, state-based planners generally conduct their search for a plan by reasoning about how the actions within a plan affect the state of the world and how the state of the world affects the applicability of actions.

This definition is in context to Artificial Intelligence. See more contextual defintions for Action-based Planning.