Keepaway-a　sub-problem　of　Robocup,　has　been　elevated　from　machine　learning　testbed　to　the　domain　to　a　machine　learning　benchmark,　which　is　much　suitable　for　comparing　different　learning　algorithms.　This　paper　introduces　the　Keepaway　Soccer　detailedly　and　compares　the　performances　of　two　RL　method-Q(λ)　and　Sarsa(λ).　The　reasons　why　they　perform　in　this　domain　are　explored　primarily.