Port-Based Modeling and Control for Efficient Bipedal Walking Robots

Research on walking robots has shown that the process of walking, in itself, requires little energy. Indeed, many robots have been built that walk with high efficiency. General analysis and control tools for such efficient walkers, however, are lacking, and many results are based on engineering intuition and ad hoc solutions. This thesis aims to provide a framework for modeling, analysis, and efficient control of walking robots. The framework uses a port-Hamiltonian system description to express the dynamics of rigid mechanisms and their interaction with the ground. The structure of the resulting models forms the basis for the development of general analysis and control techniques. The proposed framework extends well-known modeling methods to a broad class of rigid mechanisms with a configuration space described by any combination of Euclidean components (such as linear joints), Lie group/algebra components (such as ball joints), and nonholonomic components (such as nonslipping wheels). Two different 3D contact models are presented: one for compliant contact, and one for rigid contact. Using the structure of the models, the problem of finding efficient walking gaits is cast as a numerical optimization problem. This setting allows one to optimize not only the joint trajectories but also the mechanical structure of a walking robot. Finally, three control techniques for efficient walking are presented. The first technique uses the computed optimal trajectories to define a power-continuous asymptotic tracking controller. The second technique stabilizes an experimental kneed walking robot by means of a single controller on the hip joint. The third technique uses foot placement to increase the robustness of a three-dimensional walking robot.