This study aims at establishing a new control strategy for more natural and efficient bipedal locomotion. In this study, the robot is modeled as a planar biped model composed of a torso, hips, and two different legs with knees, but without ankles. The proposed method consists of Central Pattern Generator (CPG) for legged locomotion and torso attitude control. It is well known that the CPG controller copes with environmental changes by mutual entrainment of the oscillatory activities of the CPG and the body. Therefore, the biped robot can walk on both a level ground and a slope, and has the robustness for environmental changes. Moreover, the torso attitude control is executed concurrently with CPG controller for legged locomotion in the method. By utilizing the interaction between torso and legs, the biped robot with the torso can walk on the level ground over a wide range of speed. This paper presents a systematic control design method of the proposed strategy by using the genetic algorithm. In order to verify the effectiveness of the proposed method, computational simulations were carried out. As a result, it was demonstrated that the biped robot can walk on the level ground at a variable pace according to the desired torso angle given as an external command. Moreover, it was confirmed that the proposed controller has the robustness for environmental changes and external disturbance, and the biped robot can walk naturally on the uphill and downhill slopes.