Abstract : This paper addresses the problem of walk planning for a humanoid robot on a flat ground cluttered by obstacles. The algorithm we propose works in two steps. In a first step, a collision-free path for the bounding box of the robot moving in the horizontal plane is planned using classical path planning methods. In a second step, a spline is fit on the path and optimized using numerical optimization tools. The optimization criterion is time, constraints are defined by positive distance to obstacles and bounded velocities of the feet. The method has been tested on a real humanoid robot HRP-2.