With　the　rapid　development　of　the　Industrial　Internet　of　Things　(IIoT),　higher　requirements　have　been　put　forward　for　hardware　devices,　and　as　the　core　part　of　hardware　devices,　the　performance　of　chips　is　critical.　The　Obstacle-Avoiding　Rectilinear　Steiner　Minimal　Tree　(OARSMT)　problem　is　a　classic　issue　in　VLSI　physical　design,　and　having　a　shorter　OARSMT　can　help　reduce　chip　delays　and　energy　consumption.　This　paper　proposes　a　framework　based　on　Deep　Reinforcement　Learning　(DRL)　to　automatically　learn　and　generate　heuristic　algorithms　for　solving　the　OARSMT　problem.　The　core　of　our　proposed　method　combines　Graph　Convolution　Network　(GCN)　with　DRL,　where　GCN　extracts　the　graph-state　feature　information　of　Obstacle-Avoiding　Steiner　Tree　(OAST)　generated　throughout　the　process,　and　DRL　learns　policy　decisions　under　different　graph　states.　Experimental　results　have　confirmed　our　proposed　framework　outperforms　existing　manually　designed　work,　indicating　that　our　framework　is　a　promising　tool　for　solving　the　OARSMT　problem.　©　2023　IEEE.