Multi-agent Reactive Planning and Execution

The Cooperative Architecture for Plan Execution in a Mars Domain is a flexible multi-agent domain-independent architecture, which provides the agents with execution and planning capabilities in order to solve coordination problems using cooperation and negotiation during plan execution. We implemented a conflict resolution mechanism to a Mars rovers domain of NASA. In the Mars domains, Rovers are assigned plans to execute. In view of a plan failure, the conflict resolution mechanism is activated.

A cooperative architecture for a MARS domain with PlanInteraction

The proposed architecture therefore offers possibilities to model execution and planning behaviors realistically, as well as, coordination and negotiation plan execution behaviors to solve conflict.

The objective of the conflict resolution mechanism that we have implemented is two flow:

1. 1) allows the rovers to coordinate with others their actions in order to fix the plan failure with a bilateral multi-issues negotiation protocol, which use a package deal approach to negotiate multi-issues.

1. 2) If negotiation does not achieve the desired results, rovers will resort to either their on-board planning facilities for a plan repair or the central planner on Earth for a new plan.

Our contribution is a flexible MAPE architecture that allows agents to communicate, cooperate and negotiate solutions for a successful plan execution during a failure.

Relevant papers:

César Guzmán-Alvarez, Pablo Castejon, Eva Onaindia, and Jeremy Frank
Multi-agent Reactive Planning for Solving Plan Failures
8th International Conference on Hybrid Artificial Intelligent Systems, HAIS. Salamanca, Spain. Sep.11-13, (2013)