Whether a skating-training robot reproduces the prescribed motion depends on whether its servo layer can follow the motor-angle reference amplified by the screw transmission and the cable geometry. Switching active disturbance rejection control improves the large-error response by switching between linear and nonlinear corrections, but the switching frequency and the tracking accuracy lack a quantitative relation, and the hysteresis width is selected without a stated rule. This paper builds an integrated model from cable geometry and screw transmission to the permanent-magnet synchronous motor and gives the torque, speed and voltage margins of each drive; it then proposes a state-preserving hysteretic implementation in which two normalized switching variables act on the observer and feedback corrections separately, with a hysteresis band added. Six controller families are compared at fixed gains with equal tuning effort over 154 cases: the 30 rad step overshoot falls from 52.171% to 0.311%, the maximum error in the load interval falls from 0.101 rad to 0.0447 rad, and the number of mode actions falls from 46 to 18 at a hysteresis width of 0.2. With the switching frequency taken as a design variable, the servo layer can set the switching count to its target while preserving tracking quality, which offers a reusable parameterized method for the servo design of comparable cable robots.