Keeping Plan Execution Healthy

摘要

Unexpected events during the execution of a plan may lead to conflicts: we then say that the plan execution is unhealthy. This paper presents a new model that enables agents (1) to control plan-execution health and (2) to regain health when necessary. The agents can utilize the model to predict consequences of occurring disruptions and thus detect unhealthy situations. With the help of the model's predictions, agents can correct the execution of tasks within the plan to regain health. The applicability of the presented model is demonstrated by introducing two multi-agent protocols to keep the plan execution healthy. Finally, we investigate the solving capabilities and the efficiency of our method in experiments using randomly generated plans. Our conclusion is that a reasonable proportion of unhealthy situations can be solved adequately by corrections in the plan execution instead of performing a replanning procedure.