The　traditional　robotic　arm　control　method　is　based　on　the　precise　mathematical　model　of　the　task　and　lacks　adaptability.　When　the　environment　or　task　changes,　the　control　effect　is　greatly　compromised　or　even　out　of　control.　In　recent　years,　Deep　Reinforcement　Learning　(DRL),　which　has　achieved　great　success　in　games,　has　been　introduced　into　the　control　of　robotic　arms.　TD3　(Twin　Delayed　Deep　Deterministic　Policy　Gradient)　is　an　improved　algorithm　based　on　DDPG　(Deep　Deterministic　Policy　Gradient).　Like　other　DRL　algorithms,　TD3　also　has　the　problem　of　low　learning　efficiency.　This　paper　proposes　a　improved　TD3　algorithm　which　can　converge　faster　than　TD3　algorithm　in　terms　of　reachability　and　obstacle　avoidance.　Finally,　the　improvement　of　the　algorithm　is　verified　by　a　simulation　research　on　a　6-DOF　ABB-IRB1200　robotic　arm.　©　2022,　The　Author(s),　under　exclusive　license　to　Springer　Nature　Singapore　Pte　Ltd.