In　this　paper,　a　novel　reinforcement　learning　mission　supervisor　(RLMS)　with　memory　is　proposed　for　human-multi-robot　coordination　systems　(HMRCS).　The　existing　HMRCS　are　known　to　suffer　from　long　decision　waiting　time　and　large　mission　error　caused　by　repeated　human　intervention,　restricting　the　autonomy　of　multi-robot　systems.　The　proposed　supervisor　elaborately　integrates　deep-Q-network　(DQN)　and　long-short-term　memory　(LSTM)　knowledge　base　within　the　null-space　based　behavioral　control　(NSBC)　framework,　so　as　to　achieve　optimal　adjustment　strategy　of　the　behavioral　priority　in　the　presence　of　mission　conflicts,　and　to　reduce　the　frequency　of　human　intervention.　In　particular,　the　proposed　RLMS　with　memory　first　memorize　human　intervention　history　when　robot　systems　are　not　confident　in　decision　making　when　encountering　emergencies,　and　then　reload　the　history　information　when　encountering　the　same　situation　that　have　been　tackled　by　human　previously.　Simulation　demonstrates　the　effectiveness　of　proposed　RLMS　with　memory.　©　2022,　The　Author(s),　under　exclusive　license　to　Springer　Nature　Singapore　Pte　Ltd.