Dynamic Control of Legged Humanoid Robots

Over the past 20 years, control and regulation applications in robotics research have expanded from controlling fully actuated serial and parallel kinematic manipulator systems to larger, more complex systems. These include mechanical systems with non-holonomic constraints, underactuation in control inputs, and varying contact constraints, resulting in hybrid dynamic systems. Bipedal humanoid robots have become an important application area where all these control challenges arise simultaneously.

Basic walking control solutions for bipedal robots have been developed based on simplified models, but these approaches still lack robustness regarding environmental models, especially foot-ground contact, and do not generalize to the full dynamic model. Robust bipedal walking is a key capability if humanoid robots are to be seriously considered as mechanical assistants in domestic environments. Therefore, this project develops walking control algorithms that enable real-time step adaptation, allowing online adjustment to changing environmental conditions.

Compliant whole-body motion control is essential both for physical interaction with humans and autonomous manipulation. Especially for robots operating in close proximity to humans, it is necessary to master both intentional and unintentional physical interactions while standing and moving. Although modern humanoid robots can perform many controlled motions in well-defined environments, current motion control approaches mostly focus on static environments and scarcely consider physical interaction with humans. This project aims not only to enable reactive abilities to physical contact initiated by humans but also to address active human-robot interaction initiated and carried out by the robot.

The walking abilities of robots differ not only in control but also in their physical basis from those of humans. Biomechanics research has identified stable walking motions based on simple, human-inspired compliant models. While these studies focus on the fundamental principles of bipedal locomotion, applying these principles to robot control remains largely unexplored. This project aims to transfer anthropomorphic walking principles to robotic systems to achieve significant improvements in robotic walking performance. These improvements can be measured both quantitatively, such as speed or energy consumption, and qualitatively, in the ability to perform advanced locomotion like running or jumping. Implementing these principles involves developing new hardware platforms as well as control-oriented research within the Helmholtz Young Investigators Group.

A more explicit transfer of human skills to robots is particularly interesting for highly articulated humanoid robots with many degrees of freedom. Here, learning by demonstration offers the possibility to generate human-like motion behavior without relying heavily on specific optimization criteria. The learning of human motion skills in this project focuses on locomotion and force-based manipulation capabilities. Since balancing and locomotion in bipeds are inherently unstable processes, this requires extending pure imitation of human movements to an "emulation" of locomotion capabilities.

The above topics represent a control-oriented selection of motion capabilities for humanoid robots, with a focus on locomotion and physical whole-body interaction.

Publications

Book Chapters

• M. A. Roa and Ch. Ott, Balance and Posture Control for Biped Robots, Chapter 8 in Multibody System Dynamics, Robotics, and Control, X, H. Gattringer and J. Gerstmayr (Eds.), Springer Verlag, pp. 129-144, 2013. [elib]

Journal publications

• G. Garofalo, Ch. Ott, Limit Cycle Control Using Energy Function Regulation with Friction Compensation, IEEE Robotics and Automation Letters, accepted for publication. [elib]
• Ch. Ott, B. Henze, G. Hettich, T. N. Seyde, M. A. Roa, V. Lippi, T. Mergner, Comparison of bio-inspired and model-based approaches for posture control of humanoid robots, IEEE Robotics and Automation Magazine, accepted for publication. [elib]
• J. Englsberger, Ch. Ott, A. Albu-Schäffer, Three-dimensional bipedal walking control based on Divergent Component of Motion, IEEE Transactions on Robotics (TRO), accepted, 2015. [elib]
• T. Petric, L. Žlajpah, G. Garofalo and Ch. Ott, Walking with Adaptive Oscillator and Dynamic Movement Primitives, accepted for publication in International Journal of Mechanics and Control (JoMaC), 2014. [elib]
• J. Englsberger and Ch. Ott, Walking stabilization for humanoid robots based on control of the Capture Point, at-Automatisierungstechnik, Oldenbourg Wissenschaftsverlag, pp. 692-703, 11/2012.
• Ch. Ott, O. Eiberger, J. Englsberger, M. A. Roa, and A. Albu-Schäffer, Hardware and Control Concept for an Experimental Bipedal Robot with Joint Torque Sensors, Journal of the Robotics Society of Japan, Vol. 30, No. 4, pp. 378-382, May 2012. [elib]

Conference publications