Thisresearchexaminesa hybridplannerfor a real-worldmobile robotdeliveringmessagesin an office environment.The overall project, RUPART,uses "unified CBR"to combine behavior-basedcontrol withhigh-levelcase-basedplanning, usinga singlesimilaritymetricto retrievebothbehaviorcases andplan cases. Thispaperfocuseson case-basedreasoning for behaviorcases. Eachbehaviorcasedescribesa set of behaviorstargetedfor a particularenvironment.Weexplorefeatures usedto indexbehaviorcases,as wellas adaptationand learningmethods.Thisprojectis incompleteat this time, and thoroughevaluation,whileunderway,is notcomplete.