Marion Leibold + Sotiris Apostolopoulos
Institute for Automatic Control Engineering
Technische Universität München
Legged locomotion results from repeated making and dissolving contact between feet and ground. For legged robot control, hybrid models are best suited to describe the switching continuous dynamics due to changing contact situations where discontinuities from impact modeling have to be considered whenever transitions take place. The hybrid structure of the model constitutes a challenge for motion planning and control. An introduction to legged robot modeling will be given. Further, some concepts for offline optimal motion planning and control will be discussed.
Online optimal motion planning for legged robots is a computationally expensive procedure, due to the fact that the dynamic equations of motion for legged robots are not only hybrid but also non-convex and high dimensional. A way to confront this limitation is to utilize a database of precomputed control/motion primitives. In order though for a task to be adequately executed, the cardinality of such a collection has to be large, thus the primitive selection/concatenation might end up being a bottleneck for online purposes. Possible approaches include the utilization of a smaller database in combination with regression techniques to generate new primitives and the formulation of the selection problem as an MDP with a large database. When possible disturbances are not known before hand, robustness can be improved by combining primitives with invariance control. Finally continuous motion planning techniques can be utilized for unknown environments.
Marion Leibold (nee Sobotka) received her diploma in applied mathematics from Technische Universität München in 2002. In 2007 she finished her PhD on legged robot control at the Institute of Automatic Control Engineering, Universität München. She is currently a senior researcher at the Chair of Automatic Control Engineering, Universität München. Her research interests include optimal and nonlinear control theory with applications to robot control.
Sotiris Apostolopoulos is a research associate in the Chair of Automatic Control Engineering at Technische Universität München. He achieved his diploma in “Electrical and Computer Engineering” at the National Technical University of Athens (2012). His research interests include bipedal locomotion and balancing with databases of motion primitives.