The focus of this work is on sensor-based motion planning for a biped robot operating in a scene with unknown obstacles. Using on-line sensor information about its surroundings, the robot negotiates obstacles on its way. Depending on the obstacle's shape, size, and location, the decision-making system chooses an appropriate walking pattern. The robot then negotiates the obstacle and resumes stable motion. The walking pattern is chosen from a small number of precomputed patterns that together cover a reasonably wide range of possible situations. The overall control strategy is based on the concept of Zero Moment Point. Each precomputed pattern guarantees dynamically stable motion; its stability is obtained by adjusting the swing leg center of mass and hip position trajectories. The approach is fast enough for real-time implementation. Simulation experiments demonstrate stability of motion when negotiating various obstacles.