In view of the slide motion control problem of caterpillar agricultural automated robot, a sliding motion model estimation method is proposed based on the H-infinity filtering. First of all, the lateral sliding steering instantaneous centers of rotation (ICRs) motion model of the robot was given, and the double modal parallel estimation was realized for different road conditions. Then, the specific ways of implementation were analyzed according to different road conditions. Under the fixed condition, the extension H-infinity filtering (EHF) was used to estimate ICRs parameter values online in real time. Under time-varying condition, k-means clustering was used to run traffic to carry on the classification first, and EHF was used to estimate the state based on the results of classification. Simulation and experimental results show that the method under both the fixed condition and time-varying traffic can rapidly estimate the lateral sliding robot kinematics model, and keep high fitting precision of fitting.