Aimed at the problem of attitude error caused by single filtering in attitude detection of self-balancing robots, a detection strategy called “two filters and one complement” based on double filtering and neural network compensation was proposed.By analyzing the error source, an attitude compensation BP neural network for attitude sensor and attitude error was designed.According to the output information of sensor, the error compensation value of robot attitude tilt angle was predicted directly, and the attitude angle 〖JP+1〗after filtering was compensated synchronously.The compensation effect of the robot under 3 control states of standing in place, accelerating forward and backward, and turning in place was verified respectively.The results show that the attitude error of the robot can be corrected obviously by using the neural network,the neural network is helpful to improve the attitude detection accuracy of the robot