Distributed management of flexible times schedules

Abstract

We consider the problem of managing schedules in an uncertain, distributed environment. We assume a team of collaborative agents, each responsible for executing a portion of a globally pre-established schedule, but none possessing a global view of either the problem or solution. The goal is to maximize the joint quality obtained from the activities executed by all agents, given that, during execution, unexpected events will force changes to some prescribed activities and reduce the utility of executing others. We describe an agent architecture for solving this problem that couples two basic mechanisms: (1) a "flexible times" representation of the agent's schedule (using a Simple Temporal Network) and (2) an incremental rescheduling procedure. The former hedges against temporal uncertainty by allowing execution to proceed from a set of feasible solutions, and the latter acts to revise the agent's schedule when execution is forced outside of this set of solutions or when execution events reduce the expected value of this feasible solution set. Basic coordination with other agents is achieved simply by communicating schedule changes to those agents with inter-dependent activities. Then, as time permits, the core local problem solving infra-structure is used to drive an inter-agent option generation and query process, aimed at identifying opportunities for solution improvement through joint change. Using a simulator to model the environment, we compare the performance of our multi-agent system with that of an expected optimal (but non-scalable) centralized MDP solver.

Measuring the expected gain of communicating constraint information

“…Agents used information about their constraints to predict whether adding constraint information will help the team increase its utility. The significance of this work is its ability to successfully enable agents to find which types of constraints will most likely benefit from additional human information, without sharing significant information about other agent's constraints or through resource intensive queries required in other approaches [22,23,25,27,29]. In our two-stage approach, agents first initially assess which constraints will not benefit from any additional information by using our general tightness measure.…”

“…The significant in this approach is that agents reason about a limited number of interactions, allowing for tractable solutions. Previous approaches considered all possible team interactions, potentially creating a need to generate very large numbers of resource intensive "what if" queries to obtain this information [22,23,25,27].…”

“…We generated offline "what if" queries to obtain what the potential gain from information was in the latter case. It is important to stress that these "what if" queries were only used for training this machine learning model and not to be used during task executing as were done in previous approaches [22,23,25,27]. The goal of this methodology was to train a model that can estimate what types of problem instances involving quality and duration uncertainty are most likely to benefit from the user's attention.…”

Determining the value of information for collaborative multi-agent planning

“…1 Determining the value of information requires scheduling and task knowledge, because information is valuable only to the extent it influences schedule changes. In many contexts in which ASAs operate, such task and scheduling knowledge resides in a scheduler module which is external to the CI [5,42,53]. This separation of the scheduler in the systems' architecture enables other ASA modules to use the scheduler for reasoning about task schedules.…”

“…The complex, stochastic nature of changes in these settings and the pace at which they happen make it unlikely that any single individual or agent could have a complete global view of a scheduling problem [48,61]. As a result, a combination of localized reasoning and group coordination mechanisms are required for planning and scheduling decisions [53,58,4]. Autonomous agents that operate as a team, suggesting alternative courses of action to each other and negotiating to find a solution, could help people operating in such environments to achieve team objectives more effectively [27,49,43].…”