Domain-independent planning: representation and plan generation
Artificial Intelligence
A planning model with problem analysis and operator hierarchy
A planning model with problem analysis and operator hierarchy
| Hi-index | 0.00 |
This report describes a planning system that generates a sequence of operations (actions) assignments for each agent to achieve a goal statement in a multi-agent problem domain. The system is characterized by a pre-analysis of the goal statement and a representation of the breakable and unbreakable operations. The analysis would yield a sequence of parallel groups of subgoals and the execution sequence of subgoals within a group. (Two groups of subgoals are parallel to each other, if they can be achieved separately by different agents). Following this analysis result, a plan generator dynamically assigns a group of subgoals to each agent and formulates the detailed plans. In order to increase the utility of agents, the plan generator always tries to assign an idle agent to assist a busy agent if a group of subgoals is not available at that particular instance. This is done by identifying the breakable and the unbreakable subtasks of an agent and assigning one part of the breakable subtask or a sequence of unbreakable subtasks to an idle agent.Two major system components are being investigated: the goal statement analysis and the operation representations.The purpose of the goal statement analysis is to use the domain specific knowledge to examine the relationships among the subgoals, to group the correlated subgoals, to find the parallelism among the groups, and to determine the execution sequences of subgoals within a group and the pursuing orders of the groups. The system described here avoids the fruitless search efforts by 1. grouping the correlated subgoals together and assigning each group to an agent so that the complexity of a problem is reduced, 2. ordering the subgoal sequence within a group so that the trial-and-error search can be eliminated.The study of the operation representations has focused on the representations of the breakable and the unbreakable operation sequences. A sequence of operations is called breakable if each component in the sequence can be carried out by different agents. In contrast to the breakable sequence, an unbreakable sequence consists of operations that must be carried out by the same agent. The purpose of this distinction is to specify what portion of a task can be assisted by an idle agent and what portion has to be completed by the same agent if this agent has started it.