Abstract: To address the weaknesses of the traditional LQR（linear quadratic regulator）lateral control，a feed-forward-predicted LQR feedback control is proposed to realize the tracking control of roads with variable curvature.On the basis of the traditional dynamic model，the error model is built，and the feedback control quantity and the feed forward control quantity are solved respectively with the error model as the research object.To improve the adaptability of LQR to road changeability，the Q and R matrices are adjusted in real time based on the fuzzy rules which are made according to the heading and lateral position errors.Meanwhile，the road curvature information is employed to build a prediction module and update the prediction point and time in real time to address the problem of traditional LQR response hysteresis.Based on the given planning path，hardware-in-the-loop experiments are conducted，and the path tracking effects of traditional LQR and feed-forward predicted LQR under single-shift and double-shift multi-vehicle speed conditions are tested.Our results show the control effect of the newly developed predictive LQR is superior to that of traditional LQR，and the trajectory tracking error is reduced by 4.5% under single-line shift condition and 9.5% under double-line shift condition.