Robotlocomotionisanactiveresearcharea.Inthispaperwefocusonthelocomotionofquadrupedrobots.Aneffectivewalkinggaitofquadrupedrobotsismainlyconcernedwithtwokeyaspects,namelyspeedandstability.Thelargesearchspaceofpotentialparametersettingsforlegjointsmeansthathandtuningisnotfeasibleingeneral.Asaresultwalkingparametersaretypicallydeterminedusingmachinelearningtechniques.Amajorshortcomingofusingmachinelearningtechniquesisthesignificantwearandtearofrobotssincemanyparametercombinationsneedtobeevaluatedbeforeanoptimalsolutionisfound.Thispaperproposesadirectwalkinggaitlearningapproach,whichisspecificallydesignedtoreducewearandtearofrobotmotors,jointsandotherhardware.Inessenceweprovideaneffectivelearningmechanismthatleadstoasolutioninafasterconvergencetimethanpreviousalgorithms.Theresultsdemonstratethatthenewlearningalgorithmobtainsafasterconver-gencetothebestsolutionsinashortrun.Thisapproachissignificantinobtainingfasterwalkinggaitswhichwillbeusefulforawiderangeofapplicationswherespeedandstabilityareimportant.Futureworkwillextendourmethodssothatthefasterconvergencealgorithmcanbeappliedtoatwoleggedhumanoidandleadtolesswearandtearwhilststilldevelopingafastandstablegait.
