Goal-oriented vision-based walking is a fundamental high-level problem in legged locomotion. In this article, a biologically inspired cascaded control structure to realize smooth machine walking is introduced being applicable to various types of walking machines. An innovative strategy for walking pattern synthesis for adaptive biped locomotion is presented in combination with a step sequence planning algorithm. A guidance system controls the walking process based on environment perception using stereo image processing and gaze control. In the emulation environment, a set of stereo cameras mounted on a pan-tilt-head can be moved over a prototypical scenario synchronized in a feedback loop with a simulated virtual walking machine to achieve realistic perception conditions. The behavior of the controlled virtual walking machine is visualized in an augmented reality display layed over the real image of the experimental setup taken by an external camera.