``GenPlan'': Combining Genetic Programming and
Planning
EDI-INF-RR-0104. School of Informatics, University of Edingburgh, Edingburgh, UK, (December 2000)Abstract
Planning is a difficult and fundamental problem of AI.
An alternative solution to planning may lie in applying
Genetic Programming to the planning problem. As such a
Genetic Planner was constructed to assess the
feasibility of this idea. This paper introduces the
topics of Genetic Programming and Genetic Planning and
introduces the algorithm used to implement the Genetic
Planner. The Genetic Planner was applied to three
classical planning domains: STRIPS Blocks Domain,
Briefcase Domain, and the Logistics Domain. The Genetic
Planner produced good results for both the STRIPS
Blocks Domain and the Briefcase Domain. However further
work is required before it can solve any problem from
the Logistics Domain besides the trivial ones. There is
also some comparison of GenPlan with both BlackBox and
SINERGY. The first implementation provided many avenues
for further research: quick partial plan formation for
seeding the Genetic Planner's initial population, more
intelligent fitness functions, and an intelligent form
of crossover and mutation. Further research into the
feasibility of the Genetic Planner to plan in
alternative domains besides classical planning is also
important.