After a short introduction, I begin the thesis with the description of the robotic platform, where I explain my contributions in the mechanical, electrical and software design of the robots. The next part of the thesis is concerned with the stabilization methods for bipedal locomotion. I first developed a simulation platform. The control algorithms are designed based on this platform and then completed and one tuned on the real robot. My methodology facilitates rapid and robust omnidirectional walking with a velocity of over 40 cm/s for a humanoid robot of 60 cm overall height. The method is much simpler than the current state-of-the-art methods and is capable of compensating large perturbations.
The approach described here does not necessarily use accelerometers and relies on position feedback from the motors and ground contact of the feet. Afterwards, I describe several computer vision solutions I developed for the robot. The development of a color-based object recognition module is presented first. The module uses on a small low-cost CMOS camera and a low power microcontroller and provides microcontroller compatible output, in form of serial access to the list of recognized objects.
Finally, I propose two new methods for shape-based object recognition. The first method uses a grid of cells and clusters the edge points based on their orientations and reports a connection graph of the edge structure in the image. The second algorithm uses the statistics of the edge orientations in the image to find a round object using a recursive method. The ideas and methods presented in this thesis were implemented in the RoboCup humanoid team of the Free University of Berlin, the FUmanoids.