In　the　era　of　information　and　data,　the　collaboration　between　humans　and　robots　is　becoming　a　trend,　making　the　security　of　human-robot　communication　particularly　critical.　Compared　with　optical　and　electromagnetic　wave　encryption,　ultrasound　can　be　used　as　an　information　encryption　medium,　which　cannot　be　directly　captured　by　enemy　and　cannot　be　shielded　by　electromagnetic　interference.　Here,　we　creatively　present　machine　learning-enhanced　multifunctional　graphene　electronic　patches　(GEPs)　for　gesture　recognition　and　ultrasound　encryption　communication.　Thanks　to　the　multifunctionality　of　the　graphene,　GEPs　can　serve　both　as　strain　sensors　and　ultrasonic　thermoacoustic　(TA)　sources.　With　the　help　of　machine　learning,　encryption　gestures　are　analyzed　by　convolutional　neural　network　(CNN),　and　accuracy　is　as　high　as　99.3　%　and　92.37　%　at　training　set　and　test　set.　Ultrasound　robots　(URs)　controlled　by　GEPs　in　wireless　encryption　still　maintains　stable　operation　under　strong　electromagnetic　shielding.　This　work　holds　significant　application　potential　in　the　fields　of　flexible　electronics,　multifunctional　materials,　multi-robot　collaborative　operations,　and　encrypted　communication.　©　2025　Elsevier　B.V.