Abstract:
This paper presents an empirical study of some non-exhaustive approaches to optimizing preferences within the context of constraint-based, mixed-initiative planning for mission operations. This work is motivated by the experience of deploying and operating the MAPGEN (Mixed-initiative Activity Plan GENerator) system for the Mars Exploration Rover Mission. Responsiveness to the user is one of the important requirements for MAPGEN; hence, the additional computation time needed to optimize preferences must be kept within reasonable bounds. This was the primary motivation for studying non-exhaustive optimization approaches. The specific goals of the empirical study are to assess the impact on solution quality of two greedy heuristics used in MAPGEN and to assess the improvement gained by applying a linear programming optimization technique to the final solution.