Abstract: According to the requirement of the fault-tolerant control for the soft speed-sensing technology，a novel self-driving vehicle speed estimation method based on the interacting multiple-model unscented Kalman filter was proposed to adapt to the unknown statistical characteristics of the system noise． Firstly，a nominal model，which includes vehicle kinematic and dynamic characteristics，was established based on the positioning information of the self-driving vehicle，and then it was transformed into a state space nominal model including the unknown statistical characteristics of the system noise by using the forward Euler discretization method． Secondly，a series of typical values were used to describe the unknown statistical characteristics of the system noise，and a series of state space nominal models including different statistical characteristics of the system noise were obtained． For each state space nominal model，unscented Kalman filter was used to estimate the self-driving vehicle speed and all of the outputs were smoothly fused by interactive multiple-model algorithm． Thus，the interacting multiple-model unscented Kalman filter with adaptive ability to the statistical characteristics of the system noise was obtained． Simulation results show that the estimation accuracy of the proposed method for the vehicle longitudinal and lateral speeds is 4 times and 1. 5 times as many as that of the unscented Kalman filter，respectively，which satisfies the requirement of the fault-tolerant control for the self-driving vehicle．

Key words: vehicle engineering, self-driving vehicle, speed estimation, unscented Kalman filter, interactive multiple-model, adaptive