Numeric Briefcase Domain Metric Optimisation using an EA

This paper presents an extension of an evolutionary approach to classical plan optimisation to metric plan optimisation. With the advent of the third International Planning Competition (IPC), optimisation for fully automated planners is no longer solely an issue of optimising over the number of actions in a plan. The problem is more complicated and interesting as optimisation can be done in relation to a plan metric tailored for that domain and problem. The approach taken in this paper is to use an Evolutionary Algorithm (EA) to perform metric optimisation. M-GENPLAN is a system based on EA approach that performs plan optimisation as a post-processing step. This approach is compared to some current fully automated planners: Metric-FF, LPG, and MIPS. The techniques are explored using a numeric variant of the Briefcase domain, and over a small set of problems, LPG came out the winner. The evolutionary approach shows promise as it can optimise highly unfit plans, and could be used to optimise totalorder plans from other systems.