Distributed Plan and Schedule Management
Papers from the AAAI Spring Symposium
Edmund Durfee and David Musliner, Cochairs
Automated systems can be used to plan and schedule the actions and coordinated interactions of human or artificial agents. Challenges arise, however, in scaling these systems to handle increasingly complex, dynamic, stochastic, and even adversarial domains, where failure is costly. During execution, the systems need to manage unexpected deviations, and manage changes to agents’ objectives even when their plans/schedules are progressing well.
For example, when end users dynamically adopt new tasks, reprioritize existing goals, and reconsider the effectiveness of alternative actions, their associated management systems should revise their plans/schedules accordingly. If the world is unpredictable, the systems must also adapt to the environmental trajectory. Furthermore, if users’ actions can affect each other, then the systems must coordinate their evolving multiagent plans/schedules. In all of these cases, this distributed, asynchronous management must be good enough and soon enough. This symposium allowed researchers tackling aspects of the distributed plan or schedule management problem to exchange ideas, perspectives, techniques, and best practices to accelerate progress on jointly solving it.