Organizers: Jeannette Bohg, Oliver Brock
Physical contact used to be a problem in robotic manipulation research. Today, it seems to be the solution. For example, object manipulation behaviours which maximize contact and exploit contact constraints with the environment are very robust. Furthermore, physical interaction with the environment facilitates perception by creating rich, informative sensory signals that would otherwise not be present. Evidence both in humans and robots shows that these two principles of (i) exploiting contact constraints and (ii) interactive perception are essential for robust manipulation and perception under uncertainty. Yet, autonomous generation of the underlying behaviours or chaining them together remains a challenge. Existing, traditional contact models are based on a powerful and concise mathematical formalization. They make assumptions such as stable point contacts, Coulomb friction or other simplifications that ensure computational tractability. While this allows multi-contact planning, many of these assumptions do not translate well into the real world that is riddled by uncertainty.
The central question of this workshop is how we bridge between the traditional, model-based approaches and the promising contact-seeking behaviours for robust manipulation and perception. We will discuss this with researchers from different areas such as autonomous grasping and manipulation, robotic hand development, soft manipulation, whole-body control for legged robots and interactive perception. All of them share that at their core they are concerned with the problem of making and breaking contact. We aim to discuss and share the lessons learned from decades of research on manipulation of the environment via physical contact.