This　work　investigates　the　optimal　energy　allocation　considering　the　different　road　properties　for　a　series　hybrid　electric　unmanned　tracked　vehicle.　Tracked　vehicles　operate　mostly　in　off-road　conditions,　where　the　energy　consumption　changes　heavily　due　to　the　road　smoothness.　However,　few　works　considered　the　effect　of　explicit　road　properties　on　energy　allocation　for　tracked　vehicles.　Besides,　conventional　energy　management　strategies　are　generally　difficult　to　adapt　to　the　fast-changing　off-road　conditions.　To　address　these　challenges,　a　perception-guided　energy　management　strategy　based　on　deep　reinforcement　learning　that　takes　road　roughness　as　explicit　features　into　account　is　proposed.　A　method　of　road　roughness　extraction　and　quantification　is　proposed　based　on　the　random　sample　consensus　algorithm　and　singular　value　decomposition.　To　enhance　the　deployment　efficiency　in　different　off-road　driving　conditions,　a　deep　transfer　learning　framework　of　the　proposed　perception-guided　energy　management　strategy　is　devised.　Experimental　results　demonstrate　that　the　perception-guided　energy　management　strategy　improved　the　fuel　economy　by　8.15　%.　Moreover,　the　transferable　energy　management　strategy　achieves　a　convergence　rate　of　34.15　%　better　than　the　relearned　energy　management　strategy.　Our　code　is　available　at　https://github.com/BIT-XJY/PgEMS.　©　2024