I was working on making robotic manipulation and perception robust. I concentrated on multimodality and interactive perception as means to achieve robustness and to ensure that robots have access to high quality sensor data.I applied these techniques in the context of manipulation for mechanical objects, such as drawers, doors, or more complex mechanical puzzles, so called lockboxes. In order to achieve such behavior I had to become an expert in different areas of robotics, such as state estimation, sensor fusion, robot control, perception, and machine learning. Part of my research was a collaboration with behavioral biologists where we researched how Goffin's cockatoos can learn to solve mechanical puzzles. The idea here was that robots could serve as models and that we could potentially import insights from biology into robotics research.