Abstract: In WiFi-based indoor positioning technology, the stability of the RSSI (Received Signal Strength Indication)is very critical. Based on analyzing the statistical characteristics of WiFi signal strength, the truth that the RSSI does not obey the normal distribution is proved, which provides instructions for the construction of stochastic model of filtering. By integrating Friis transmission equation and the kinematic equation into robust kalman filter method, we can reach the target of smoothing the signal, which can ensure the accuracy of WiFi indoor positioning from the signal source. According to the research results, the average positioning error in static applications can be reduced to 0.3 meters compare with the signal attenuation model, while for dynamic applications, it has a much better performance than the signal attenuation model, the average positioning error can be reduced from 5 meters to 1.4 meters.