Welding robot is a vital part of automatic welding process. At present, in domestic, robot path planning relies on human experience mostly. Repeated debug is needed later, resulting in increased workload. To solve this problem, an improved ACO algorithm was proposed, and a path with the least interference and the shortest time in the welding process was given. The welding simulation experiment was performed in ROBCAD, considering the accessibility and interference problem of the robot. Lastly the simulation time was calculated. The improved ACO algorithm has improved the welding efficiency and welding time, which provides reference for further study of robot welding path.