In　order　to　improve　the　safe　interaction　between　the　robot　and　the　external　environment,　a　tactile　sensor　model　for　the　curved　surface　of　the　robot　was　proposed　based　on　the　theory　of　electric　potential　distribution　on　the　conductive　surface　and　the　closed　gas　compression　law.　The　sensor　might　detect　contact　positions　and　contact　pressures.　Three-layer　structure　was　adopted,　including　conductive　layer,　compressed　gas　isolation　layer　and　signal　extraction　layer.　In　order　to　adapt　to　arbitrary　surface　shapes　and　reduce　the　influences　of　nonlinear　electric　potential　distribution　of　conductive　layer,　a　machine　learning　algorithm　was　used　to　reconstruct　the　electric　potential　distribution　model.　The　conductive　layer　of　the　sensor　was　modeled　and　simulated　by　COMSOL　software,　and　the　sensor　sample　was　prepared.　Simulation　and　experimental　results　show　that　the　proposed　tactile　sensor　may　be　customized　to　the　curved　surfaces　of　the　robot,　and　may　realize　the　real-time　detection　of　contact　positions　and　contact　pressures,　which　may　be　used　for　human-machine　information　interaction.　©　2022　China　Mechanical　Engineering　Magazine　Office.　All　rights　reserved.