Distributed approaches for solving constraint-based multiagent scheduling problems

• Authors: Edmund H. Durfee;James C. Boerkoel, Jr.
• Venue: Distributed approaches for solving constraint-based multiagent scheduling problems
• Year: 2012

This research focuses on building foundational algorithms for scheduling agents that assist people in managing their activities in environments in which tempo and complexity outstrip people's cognitive capacity. The critical challenge is that, as individuals decide how to act on their scheduling goals, scheduling agents should answer queries regarding the events in their interacting schedules while respecting individual privacy and autonomy to the extent possible. I formally define both the Multiagent Simple Temporal Problem (MaSTP) and Multiagent Disjunctive Temporal Problem (MaDTP) for naturally capturing and reasoning over the distributed but interconnected scheduling problems of multiple individuals. My hypothesis is that combining a bottom-up approach—where an agent externalizes constraints that compactly summarize how its local subproblem affects other agents' subproblems, with a top-down approach—where an agent proactively constructs and internalizes new local constraints that decouple its subproblem from others', will lead to effective solution techniques. I confirm that my hypothesized approach leads to distributed algorithms that calculate summaries of the joint solution space for multiagent scheduling problems, without centralizing or otherwise redistributing the problems. In both the MaSTP and MaDTP domains, the distributed algorithms permit concurrent execution for significant speedup over current art, and also increase the level of privacy and independence in individual agent reasoning. These algorithms are most advantageous for problems where interactions between the agents are sparse compared to the complexity of agents' individual scheduling problems. Moreover, despite the combinatorially-large and unwieldy nature of the MaDTP solution space, I show that agents can use influence spaces, which compactly capture the impact of agents' interacting schedules, to tractably converge on distributed summaries of the joint solution space. Finally, I apply the same basic principle to the Hybrid Scheduling Problem, which combines constraint-based scheduling with a rudimentary level of planning, and show that my Hybrid Constraint Tightening precompilation algorithm can improve the propagation of information between planning and scheduling subproblems, leading to significant search space pruning and execution time reduction.