It is proposed the Local Axis Gait algorithm in order to generate real-time gait patterns for a humanoid robot. The 3D foot motion planning for the humanoid global motion is developed in order to walk in any surface as plane, ramp, climbing stairs. Furthermore, it is possible continuous change the step length and orientation in real time. The cart-table model is used for planning COG and ZMP motion. The proposed algorithm takes into account physical robot constraints as joint angles, angular velocity and torques. Torques are computed by Lagrange method under Lie groups. Some results are shown and discussed. © 2008 IEEE.