With the miniaturization of medical equipment and the comfort of diagnosis mode, positioning has become a major challenge in the research of capsule robots. In order to realize the effective positioning of magnetically controlled capsule robots, a positioning method based on double-layer symmetric sensor array is proposed in this work. Firstly, a near-field error correction method based on multi-magnetic dipole model is presented to improve the accuracy of capsule robot magnetic field model. Then, for multi-magnetic driving environment, an arrangement of double-layer symmetric sensor array is designed to reduce the interference of hybrid magnetic fields. The effectiveness of this arrangement is verified by calculating the elimination error of magnetic field in symmetric planes. Driving and positioning device are developed, and a BP neural network model is trained by recording the specific position of the capsule robot and the corresponding magnetic field. The trained BP neural network model can provide a precise initial solution to improve the ability of LM algorithm to solve the position of the capsule robot. Finally, the capsule robot is placed in a porcine stomach in vitro to test the positioning method by gauge points, and the average positioning error is 3.4 mm. The proposed positioning method has high accuracy and meets the needs of clinical inspection of capsule robot.