To address the problems of end-effectors navigation efficiency, real-time performance, robustness, and path global optimization of a dual-arm service robot, a dual-arm service robot end-effectors path planning method is proposed based on the improved rapidly-exploring random trees algorithm. The method uses random sampling of two random tree parent node links, combines the target deviation angle, and random values to change the fixed-step search strategy, and introduces the artificial potential field method to locally optimize the random sampling, effectively balancing the randomness and blindness of the original algorithm, thus improving the path quality and shortening the planning time. After that, the path redundant points are removed and the path is smoothed by a cubic B-spline curve to optimize the end motion of the dual arms and reduce jitter. The master-slave planning method is used to plan the obstacle avoidance of the master arm first. Then, the slave arm plans the obstacle avoidance and collision avoidance paths according to the path of the master arm. Through MATLAB simulation and a real experimental platform, it evaluates that the algorithm outperforms the traditional RRT and other improved algorithms in terms of the number of iterations, the planning time, and the final path length under an environment of the same complexity. It significantly improves the efficiency and quality of the path planning of the dual-arm service robot.