Under　denial-of-service　(DoS)　attacks,　achieving　a　desired　formation　of　networked　nonholonomic　mobile　robots　(NMRs)　over　communication　is　a　nontrivial　challenge　due　to　communication　disconnecting　and　recovering　at　any　time.　Efforts　have　been　made　in　prevention,　detection,　and　resilience　methods,　but　existing　methods＇　generalization　ability　and　interpretability　still　need　improvement.　This　paper　proposes　the　leader-following　formation　control　framework　for　a　group　of　NMRs　under　denial-of-service　(DoS)　attacks.　The　Koopman　operator　is　utilized　to　express　a　linear　lifted　feature　space　for　the　discrete　motion　of　networked　NMRs.　Based　on　an　event-triggered　mechanism,　a　data-driven　loss　function　is　presented　to　approximate　the　Koopman　operator,　enabling　a　long-term　recovery　capability　under　DoS　attacks.　Furthermore,　according　to　the　recovered　motion　of　the　leader,　the　leader-following　formation　controller　with　a　variable　gain　is　designed　to　ensure　that　the　formation　error　converges　to　the　neighborhoods　of　zero.　In　numerical　simulation,　the　effectiveness　of　the　proposed　method　is　verified　by　a　DoS　attack　example.　©　2023　IEEE.