Autonomous　behavior　motion　planning　with　kinematics　constrains　is　improved　for　lunar　rovers.　The　behavior　controller　is　implemented　by　fuzzy　system,　and　the　hybrid　architecture　is　adopted　to　ensure　both　the　robustness　and　optimization.　Velocity　and　turning　radius　are　chosen　as　the　output　of　fuzzy　behavior　to　decouple　the　velocity　and　angle　velocity.　Behavior　decision　field　(BDF)　is　presented,　and　kinematics　constrains　are　re-written　as　the　reward　functions　within　the　Q-learning　process.　This　bottom-to-top　design　ensures　motion　planning　the　kinematics　quality　in　BDF.　Experiment　results　are　shown　to　demonstrate　the　effectiveness　and　traceability　of　the　trajectory.