Recently,　the　rise　of　the　virtual　reality　(VR)　and　augmented　reality　(AR)　application　has　led　the　significant　evolution　of　smart　Internet　of　Things　(IoT).　To　improve　the　transmission　efficiency　and　system　overheads　for　IoT　system,　edge　computing　and　caching　have　been　treated　as　a　prospective　technology.　However,　there　are　some　inevitable　issues　need　to　be　emphasized　such　as　limited　computing　and　caching　resources,　as　well　as　centralized　learning　and　training　by　agents　for　resource　allocation.　In　order　to　tackle　above　issues,　in　this　article,　based　on　deep　reinforcement　learning　(DRL),　we　propose　and　introduce　a　novel　collective　and　intelligent　optimization　algorithm　-　collective　deep　reinforcement　learning　(CDRL).　Through　collective　learning　and　training　by　multi-agents,　it　can　avoid　excessive　consumption　of　energy　resources,　and　realize　intelligent　resource　allocation.　Moreover,　blockchain　technology　is　viewed　to　guarantee　the　security　of　shared　data　for　the　cloud-edge-end　collaborative　IoT　network.　optimized　design　for　the　offloading　decision　of　computation　tasks　of　VR　device　and　the　consensus　task　of　blockchain,　as　well　as　caching　decision,　the　computing　overheads　and　energy　consumption　aim　to　be　minimized.　Then,　we　design　the　formulated　problem　as　a　Markov　decision　process,　the　optimal　scheduling　and　policies　can　be　achieved　based　on　the　CDRL.　Simulation　results　reveal　that　the　proposed　scheme　outperforms　other　existing　schemes.　©　2023　IEEE.