Sensor-based robust whole-body control of highly dynamic legged humanoid robots




Kim, Donghyun, Ph. D.

Journal Title

Journal ISSN

Volume Title



Industrial robots significantly improve the productivity of manufacturing operations performing various tasks rapidly, accurately, and repeatedly. It would be hard to imagine factories without robotic arms. At the same time, it is difficult to imagine human-centered robots maintaining infrastructure and providing care as they are not yet versatile enough. One important obstacle to the adoption of human-centered robots is their limited mobility. Legged humanoid robots represent an embodiment of a highly dexterous system which could provide human-like capabilities to boost automated services in human environments. Therefore this thesis is dedicated to investigate the sensor-based control of legged humanoids robots such that they can achieve versatile and high task performance. To tackle agility and robustness in legged humanoid robots, I have studied the dynamic whole-body motion control of these kind of robots, with special focus on dynamic locomotion in coordination with whole-body task capabilities. One of the unique aspects of this study is the enhancement of locomotion capabilities without compromising the robot’s dexterity. Currently, existing locomotion techniques for legged systems are highly specialized and not adaptable to generic robotic structures with manipulation requirements. Here, we explore the robot’s legged mobility without compromising its dexterity by utilizing a general-purpose whole-body controller (WBC), i.e. a control algorithm which can find a dynamically-consistent mapping from operational space tasks to joint torques. The use of a WBC is appealing due to its ability to coordinate multiple tasks for highly redundant robotic systems. As such, WBCs have been deployed recently for controlling humanoid robots. However, the use of WBCs for achieving highly dynamic sensor-based motions has been lacking, and our work addresses the technical problems of such and endeavor. Our research primarily focuses on employing WBCs for dynamic motion control of legged robots. The dynamic control of real robots requires both algorithmic developments and compre- hensive system analyses for real-time deployment, which covers a broad spectrum of components from motor-level control to high-level planners. Therefore, my studies include the algorithmic enhancement of WBCs, the development of locomotion planners, the analysis of real-time con- trollers, and the integration of state-estimators. The algorithmic theory and methods are verified in both simulation and various real systems.


LCSH Subject Headings