Two ways I’ve been training locomotion policies :
Goal Base : Moving to randomized waypoints

Controls Based : Direction and speed input commands