Aimed at the timevarying perturbation caused by wheelset tread radial deformation squeezing locomotive brake cylinder piston and the brake shoe in actual working conditions, a supertwisting observer (STO) was designed. The actual value was estimated, and the estimation error was approached by using adaptive fuzzy approximation method to weaken the timevarying perturbation. An adaptive robust controller based on differential control law (DCL) was constructed and applied to the position control system of brake shoe; the controller parameters were optimized by using the gravitation algorithm (GSA) with chaos variation factor. Theoretical analysis and simulation results show that the output of each state variable in the closedloop system is bounded, the derivative of the sliding mode surface converges to 0 and the sliding mode surface is constantly converged; control input is smooth bounded and no chattering.