Biped robot control techniques are usually based on the tracking of pre-computed reference trajectories. Therefore, to achieve autonomy in locomotion, it is necessary to store a set of trajectories handling all the possible situations and events the system is susceptible to encounter. This paper presents a method for the generation of walking gaits for biped robots. Theses gaits define locomotion in the three dimensional space (3D) and are parameterized with the steps characteristics. The approach we propose is based on an actuator input minimization under constraints. These constraints are linked to the gait parameters (step length, step clearance...). The method is based on a simple and intuitive reasoning, allowing the designer to define walking trajectories in a fast and easy way. We present simulation results obtained with the 15 actuated joints biped robot of INRIA called BIP.