Abstract: To overcome the limitations of traditional single-arm welding robots in adapting to complex postures and variable-trajectory operations, a dual-arm humanoid welding robot system was proposed. This system can perform welder-like operations, with its left arm delivering welding wire and its right arm performing the welding operation. A control strategy based on inverse kinematics was adopted to achieve anthropomorphic motion and prevent collisions caused by extreme joint configurations. Based on manually input weld start and end positions and preset tool poses, a joint configuration method based on damped least squares was proposed, and dual-arm motion trajectories were generated. Experiments involving flat welding, butt welding, and patch welding of 304 stainless steel, as well as flat welding of Q355 steel, were conducted under the conditions of fixed workpieces and preset process parameters. The feasibility of the developed system in performing dual-arm collaborative tungsten inert gas (TIG) welding tasks along preset trajectories was verified by the experiments. The results indicate that, for the 12 sets of target poses listed, the average norm of the position residual obtained from the inverse kinematics solution is 0.091 mm. This provides a viable scientific method and technical approach for advancing welding intelligence through the use of humanoid welding robots.