Robust Bipedal Locomotion over Rough Terrain by extending ZMP-based Control

Humanoid robotics is a very active and recent research field, that aims at developing robots which are suitable to interact with an environment designed for humans. One of the biggest challenges is the generation and control of a stable dynamic biped locomotion on any terrain. In order to ensure the dynamic stability, the Zero-Moment Point (ZMP) criterion has been widely used. Based on it, the ZMPPreview Control proposed by Kajita et al. has become the most used walking approach for bipedal robots. Although the basic method focuses on generating a stable horizontal motion for walking on flat ground, various extensions have then been proposed for uneven terrain scenarios which are discussed in this work. A basic ZMP-Preview Control limited to walking on flat ground is already implemented on the humanoid robot “Johnny #5”, which is used as experimental platform in this work. Hence, this thesis investigates how to extend this existing software to enable walking motions on uneven terrain. The application of a unique method using Virtual Slopes shows first promising results and was successfully tested in simulation on uneven terrain such as stairs. On the real robot, a robust ZMP Balance Control is additionally required. For this reason, the already existing approach was re-designed by combining state of the art balance controllers. In order to evaluate and tune the controller performance, visualization tools are provided by the newly implemented software stack as well