EASE

Overview

In the DFG Collaborative Research Center Everyday Activity in Science and Engineering (EASE), we study how robots could perform everyday tasks - such as cooking - with the same skill as humans. As an interdisciplinary research project, EASE explores many facets of human and robotic movement processes and problem-solving strategies in the household domain. Our research group is home to scientists from subprojects P01 and P05, working on a variety of research questions:

• Can language be better understood using simulation?
• How can the robot learn the affordances of available objects?
• How can the robot's memories be represented (e.g. with the help of ontologies)?
• What principles of human metacognition ("thinking about thinking") can help the robot plan, execute, and troubleshoot actions?
• How might different, hybrid systems that support the robot (neural networks, logic control systems, physics simulation, etc.) communicate with each other, and how can their outputs be combined to produce results that are useful to the robot?
• Can we anticipate the robot's queries in order to predict answers and have them readily available without reaction time?

In close collaboration with other research groups and subprojects, we develop different approaches to answer these questions. For example, we are programming a hybrid query engine that coordinates the different information systems. The logged queries plus execution context shall then be used for anticipation using machine learning techniques.
In addition, we model the robot's internal control systems and their workflows so that the robot can get an idea of its own thought processes and learn to control them autonomously. For example, if the robot fails to grip a cup correctly, we want it to retrace the planning process to identify the cause of the error and possible solutions: If the position of the cup was perceived incorrectly, it should take another look at it from a different perspective. If a joint is jammed, he better uses the other arm.