Task Scheduling using Constraint Optimization with Uncertainty

Multiagent task scheduling encompasses diverse domains of problems that require complex models and robust solutions. C-TÆMS [1] is a new specification, based on TÆMS [2], for multiagent task scheduling problems that represents the complex relationships necessary to model these domains. Some recent work has been done in the area of mapping the C-TÆMS scheduling problem into a Distributed Constraint Optimization Problem (DCOP) [6]. Distributed constraint optimization is a direct extension to the traditional AI approach of constraint satisfaction for multi-valued constraints in a distributed system [7, 3]. Typical DCOP algorithms define the optimal solution as the optimal sum of local utilities. Currently the mapping from C-TÆMS to a DCOP allows only for certain combinations of quality accumulation functions (QAFs), and works only for deterministic outcomes. The C-TÆMS scheduling problem contains uncertain information describing possible outcome distributions over the qualities of methods. The combination of these possible outcome distributions creates uncertainty in the global utility of a task schedule. Using an evaluation function for comparisons, the optimal schedule may not be equal to the one with the optimal sum of local utilities. In this paper we extend the original DCOP formalization for uncertainty information in the form of utility distributions. Additionally, we extend the C-TÆMS mapping to include additional QAF combinations using only binary constraints. We then show how the C-TÆMS mappings can take advantage of the extended DCOP formalization with some sample evaluation functions. This research is ongoing, and comprehensive results on general classes of C-TÆMS scheduling problems are pending.