The goal of this work is to provide building blocks for resource-aware robot architectures. The topic of these blocks are data-driven generation of context-sensitive resource models, prediction of future resource utilizations, and resource-aware computer vision and motion planning algorithms. The implementation of these algorithms is based on resource-aware concepts and methodologies originating from the Transregional Collaborative Research Center "Invasive Computing" (SFB/TR 89).

Diese Arbeit präsentiert einen Ansatz zur datengetriebenen Bewegungsgenerierung für humanoide Roboter, der auf der Beobachtung und Analyse menschlicher Ganzkörperbewegungen beruht. Hierzu wird untersucht, wie erfasste Bewegungen repräsentiert, klassifiziert und in einer großskaligen Bewegungsdatenbank organisiert werden können. Die statistische Modellierung der Transitionen zwischen charakteristischen Ganzkörperposen ermöglicht im Anschluss die Generierung von Multi-Kontakt-Bewegungen. This work presents an approach to data-driven motion generation for humanoid robots, which is based...

Equipping robots with complex capabilities still requires a great amount of effort. In this work, a novel approach is proposed to understand, to represent and to execute object manipulation tasks learned from observation by combining methods of data analysis, graphical modeling and artificial intelligence. Employing this approach enables robots to reason about how to solve tasks in dynamic environments and to adapt to unseen situations.

The goal of this work is the development of a novel computational formalization of whole-body affordances which is suitable for the multimodal detection and validation of interaction possibilities in unknown environments. The hierarchical framework allows the consistent fusion of affordance-related evidence and can be utilized for realizing shared autonomous control of humanoid robots. The affordance formalization is evaluated in several experiments in simulation and on real humanoid robots.

Robustly maintaining balance on two legs is an important challenge for humanoid robots. The work presented in this book represents a contribution to this area. It investigates efficient methods for the decision-making from internal sensors about whether and where to step, several improvements to efficient whole-body postural balancing methods, and proposes and evaluates a novel method for efficient recovery step generation, leveraging human examples and simulation-based reinforcement learning.

Ein augmentierendes Exoskelett wird bestenfalls erst vom Benutzer wahrgenommen, wenn es unterstützende Drehmomente auf die Gelenkachsen des Benutzers aufprägt. Die vorliegende Arbeit befasst sich deshalb mit der Entwicklung von kinematisch kompatiblen Gelenkmechanismen für Beinexoskelette und deren Evaluierung über die Interaktionskräfte an der Mensch-Roboter-Schnittstelle. Basierend auf diesen Interaktionskräften wird zudem ein Ansatz zur Online-Klassifikation von 13 Bewegungstypen vorgestellt. Exoskeletons for augmentation should be transparent until assistive torques on the user's...

Visual perception is one of the most important sources of information for both humans and robots. A particular challenge is the acquisition and interpretation of complex unstructured scenes. This work contributes to active vision for humanoid robots. A semantic model of the scene is created, which is extended by successively changing the robot's view in order to explore interaction possibilities of the scene.