In　this　paper,　a　neural-network-based　policy　learning　method　is　established　to　solve　robust　stabilization　for　a　class　of　continuous-time　nonlinear　systems　with　both　internal　dynamic　uncertainties　and　input　matrix　uncertainties.　First,　the　robust　stabilization　problem　is　converted　to　an　optimal　control　problem　by　choosing　an　appropriate　cast　function　and　proving　system　stability.　Then,　in　order　to　solve　the　Hamilton-Jacobi-Bellman　equation,　a　policy　iteration　algorithm　is　employed　by　constructing　and　training　a　critic　neural　network.　The　approximate　optimal　control　policy　can　be　obtained　by　this　algorithm,　and　the　solution　of　the　robust　stabilization　can　he　derived　as　well.　Finally,　a　numerical　example　and　an　experimental　simulation　are　provided　to　verify　the　availability　of　the　proposed　strategy.