Improving a Plan Library for Real-Time Systems Using Nearly Orthogonal Latin Hypercube Sampling

Abstract

Computing solutions to intractable planning problems is par-ticularly problematic within real-time domains. One ap-proach to this challenge includes off-line computation, such as precomputing a plan library. However, because complex domains preclude creating a comprehensive library, a system must choose a subset of all possible plans to include in the library. Strategic selections will reduce the probability that a system encounters a situation for which it does not have an appropriate plan in the library to either apply directly or adapt. Choosing variable values using Latin hypercubes is a tech-nique used to reduce the number of test cases required in or-der to validate complex systems. Here we discuss the applica-tion of a variation of this technique, nearly orthogonal Latin hypercubes, to planning spaces in order reduce the number of plans a system must cache in its library.