LearningtoCooperateusingtwoSix{Legged

Abstract:  leggedautonomousmobi lerobotsperformingac omplextask.Therobots arecapable ofcommunicatingwithe achotherviaaradiolin k.Therobottaskconsis tsofeciently movinganobjecttoagoa llocation,andrequire scooperationofthetwo robots.We usethistasktodemonst rateanapproachtoconc urrentlearninginarea l{worldwith complexdynamicsofint eractionbetweenthero botsthemselves,andth erobotsand theobject.Inonesetof experimentswedemonst ratetheperformanceof theproposed learningapproachonth eproblemofconcurrent lylearning\whattodo& quot;and\whatto communicate".In anothersetofexperime ntswecomparetwodiere ntreinforcement strategiesandshowthe ireectivenessonthepr oblemoflearning\what tocommuni-cate." ; KeyWords.reinforceme ntlearning,robotics, autonomousagents,mul ti{agentsys-tems,rea l{timeissues,walking robots.