With　the　rapid　development　of　driverless　technology,　supporting　the　safe　and　rapid　path　planning　of　Unmanned　Ground　Vehicles　in　complex　environments　turns　to　be　a　significant　challenge.　In　this　paper,　we　propose　a　3D　environment-based　multiobjective　path　planning　strategy　with　consideration　of　the　environmental　characteristics　and　time　cost　By　using　the　deep　reinforcement　learning　algorithm,　we　further　transform　the　initial　problem　into　Markov　decision　process　which　can　be　solved.　We　consider　grid　quantization　approach　where　the　3D　terrain　of　the　environment　surface　can　be　converted　into　a　height-based　hierarchical　matrix　and　feed　the　matrix　into　the　reinforcement　learning　architecture　as　the　information.　Meanwhile,　we　propose　an　action　judgement　mechanism　to　judge　the　legitimacy　of　actions　in　advance　before　execution　to　solve　the　collision　problem　in　the　real　training　process　of　the　agent.　Simulation　results　show　the　effectiveness　and　robustness　of　the　proposed　strategy　with　applications　to　two　different　types　of　3D　terrains　and　varying　degrees　of　terrain　disturbances.　©　2022　IEEE.