With　the　widespread　application　of　rehabilitation　robots,　surface　electromyography　(sEMG),　as　a　non-invasive　way,　has　become　an　effective　tool　to　judge　the　activation　state　of　the　related　muscles.　The　traditional　muscle　activation　model　is　not　accurate　enough　for　the　reason　that　the　sEMG　signals　contain　wealth　of　unknown　information.　To　estimate　the　muscle　activation　state　and　to　evaluate　the　effect　of　ankle　rehabilitation　training,　the　activation　mean　sample　entropy　is　proposed　in　this　paper　which　is　modified　by　the　sample　entropy　of　the　complexity　of　the　expression　sequence　based　on　the　traditional　muscle　activation　model.　The　degree　of　muscle　activation　in　the　process　of　ankle　rehabilitation　training　was　evaluated　by　using　a　parallel　ankle　robotic　system,　and　the　accuracy　of　the　index　was　verified　by　conducting　experiments　with　a　healthy　subject.　©　2021　IEEE.