Graduation Semester and Year




Document Type


Degree Name

Master of Science in Computer Engineering


Computer Science and Engineering

First Advisor

Manfred Huber


Despite significant progress in recent years, the capabilities of today's Humanoid robots still lack behind the walking abilities of humans in terms of competence, robustness, flexibility, and speed. Furthermore, unknown environmental conditions and related constraints imposed on the robot significantly increase the complexity of locomotion control and decision making for such systems, easily making planning-based approaches intractable. Part of the reason for this is that the control of a humanoid encompasses observation, processing the observed data, and decision making in terms of locomotion gait and body pose parameters. This key information derived from observations and internal robot information finally allows calculating the appropriate signals to be fed to the actuators which in turn move and adjust the mechanical joints with respect to the environment. The complexity of biped walking is also driven by the kinematic structure of the robot. If the robot has a large degree of freedom, the parameters that can be used to affect the robustness of the robot increase along with the number of controls. This, in turn, can lead to a significant increase in computation cost in monolithic control approaches that compute gait and control for the entire kinematic mechanism. As a contribution towards the objective of developing useful walking machines, the work presented in this thesis takes a modular approach to locomotion control where the overall control task is decomposed into elements with individual subtask responsibilities. The goal here is to break the overall complexity into manageable parts by relying on the robustness and reactivity of the other modules. This thesis presents a basic overview of this approach and then focuses on the development of the parts of this approach centered around flexible gait generation. In this part it focuses on modules that address very specific problems of walking such as permitting dynamically changing step lengths, stepping frequencies, height of the body, and stance stability during the walk cycle, in order to adjust itself to the environment, prevent it from falling down, and address foothold and pace parameters provided by higher-level, environment-dependent modules. This thesis proposes a control framework that stabilizes a humanoid robot while these characteristics of the walk are changed. In the modules developed to achieve this, methods such as position control, flexible walking pattern generation using parametric trajectories, and zero-moment control for reactive stabilization are used to generate a dynamically walk. The resulting controller is demonstrated using a simulated humanoid model taking into account the natural dynamics, torque limits and the model of the walking surface.


Computer Sciences | Physical Sciences and Mathematics


Degree granted by The University of Texas at Arlington