A framework for compliant physical interaction

Abstract

Although the grasp-task interplay in our daily life is unquestionable, very little research has addressed this problem in robotics. In order to fill the gap between the grasp and the task, we adopt the most successful approaches to grasp and task specification, and extend them with additional elements that allow to define a grasp-task link. We propose a global sensor-based framework for the specification and robust control of physical interaction tasks, where the grasp and the task are jointly considered on the basis of the task frame formalism and the knowledge-based approach to grasping. A physical interaction task planner is also presented, based on the new concept of task-oriented hand preshapes. The planner focuses on manipulation of articulated parts in home environments, and is able to specify automatically all the elements of a physical interaction task required by the proposed framework. Finally, several applications are described, showing the versatility of the proposed approach, and its suitability for the fast implementation of robust physical interaction tasks in very different robotic systems.

Reliable non-prehensile door opening through the combination of vision, tactile and force feedback

“…The hand frame, H was set to the inner part of the robot fingertips, whereas the grasp frame, G, was set to the handle, according to our previous work on a framework for specifying physical interaction tasks (Prats et al, 2010).…”

“…Finally, frames H, G and T denote the hand, grasp and task frames, and are given with respect to R and O respectively. They are used as auxiliary entities for grasp and task description and execution (Prats et al, 2010). …”

“…We assume that a task-oriented grasp planning algorithm, based on our previous work (Prats et al, 2007b(Prats et al, , 2010, exists, providing the following information:…”

Manipulation primitives: A paradigm for abstraction and execution of grasping and manipulation tasks

“…Despite different definitions of primitives, all of them present a common trend, discretizing and reducing the complexity of controlling a robotic setup by reducing the search space for planning. Other similar approaches include Object Action Complexes [14] and the physical interaction framework of [15]. However, in contrast to this work, all of the above consider primitives which are specific to a particular embodiment.…”

Robot learning of everyday object manipulations via human demonstration

“…at logic reasoning layer). A notable exception is the work of Prats et al [20]. However, there is still a gap between these two different levels.…”