The goal of the presented work is to use a low-dimensional control interface to control the high-DOF whole-body locomotion motion of a humanoid robot, for instance using a 2D keyboard or joystick interface (up-down, left-right) to control a 30 DOF acyclic locomotion task (reaching a goal position by climbing on a stair). This work is motivated by the use of non-invasive BCI, which offers only such low-dimensional control signals, for low-level motion control of humanoids or exoskeletons in the assistive robotics domain of applications. The methodology is the following: given the target complex locomotion task, the humanoid autonomously plans the high-DOF motion and then executes it allowing the user to control on-line some low-dimensional features of the motion, namely way-points of the moving end-links. The approach is based on the two-stage contact-before-motion planning paradigm, which autonomously plans a sequence of transition contacts in its first stage, then executes the collision-free dynamics-consistent motion in the second stage, keeping dynamics balance of the motion. Example of this control approach is demonstrated in dynamic simulation.