A vision for the future are autonomous and semi-autonomous systems that perform complex tasks safely and robustly in interaction with humans and the environment. The action of such a system needs to be carefully planned and executed, taking into account the available sensory feedback and knowledge about the environment. Many of the existing approaches view motion planning as a geometrical problem, not taking the uncertainty into account. Our goal is to study how different type of representations and algorithms from the area of machine learning and classical mathematics can be used to solve some of the open problems in the area of action recognition and action generation.FLEXBOT will explore how how topological representations can be used for an integrated approach toward i) vision based understanding of complex human hand motion, ii) mapping and control of robotics hands and iii) integrating the topological representations with models for high-level task encoding and planning.Our research opens for new and important areas scientifically and technologically. Scientifically, we push for new way of thinking in an area that has traditionally been born from mechanical modeling of bodies. Technologically, we will provide methods plausible for evaluation of new designs of robotic and prosthetic hands. Further development of machine learning and computer vision methods will allow for scene understanding that goes beyond the assumption of worlds of rigid bodies, including complex objects such as hands.