A Compliant Multi-finger Grasp Approach Control Strategy Based on the Virtual Spring Framework

Abstract

This paper presents an adaptive compliant multi-finger grasp approach control strategy based on based on a new interpretation of the virtual spatial spring framework, to improve the grasp performance for target objects with position errors. An n-finger virtual spatial spring frame is proposed to achieve the adaptive compliant grasp control. Twofinger grasp control based on a single virtual spring is tackled, and then extended to multi-finger grasp control. Virtual springs for self-collision avoidance among digits are constructed to form the complete adaptive compliant grasp control law. With the virtual-spring based adaptive compliant grasp approach control strategy, the first robot finger to experience unexpected impact remains in contact with the object, while the rest of the fingers are continuously, adaptively driven toward readjusted grasping positions by the virtual springs without the need for on-line replanning. Experimental results demonstrate effectiveness of the virtual-spring based grasp controller, and significantly larger position errors of the target object can be accommodated with the proposed adaptive compliant grasp control strategy.

Haptic Exploration of Unknown Objects for Robust In-Hand Manipulation

“…Recent works emphasise the importance of physical compliance by achieving complex in-hand manipulation using soft hands [55], [56]. Where such hardware is not available, a common way to manipulate unknown objects is the VSF, that is, creating a virtual coupling between the robot fingers to apply the grasping forces [7], [16]- [18]. In grasping, it is shown to increase robustness under uncertainty [18].…”

“…We execute the tasks that are learned from demonstration using a four-finger anthropomorphic robot hand, on objects of different shapes and surface materials. We compare our exploration-based approach to our baseline, the VSF, which is widely used for object-agnostic grasp and manipulation [7], [15]- [18], including our previous work [5]. Our current work differs from [5] as we add a haptic exploration step before the manipulation and apply GFO to maintain a better grasp.…”

Learning by Demonstration and Robust Control of Dexterous In-Hand Robotic Manipulation Skills

“…In this work, we use a virtual springs approach for the coordination between multiple fingers. It has been already applied to grasping [2] and dexterous manipulation problems ( [21], [9]). In our approach, we do not modify the DMP formulation, but create a coupling at a lower level of control.…”

“…In our approach, we do not modify the DMP formulation, but create a coupling at a lower level of control. Virtual springs also add compliance to the system as shown in [2]. This choice also allows us to learn object-centric trajectories, hence the task-level control is independent of the underlying robot hardware.…”

“…However, the application of DMPs to in-hand manipulation is not trivial, and to our knowledge has not been attempted so far, as it requires the coordinated control of the trajectory of multiple fingers, which have to maintain contacts and to exert appropriate forces on the manipulated object simultaneously. In order to satisfy these requirements, we use a compliant controller, based on the virtual springs framework (VSF [21], [2], [9]), which takes advantage from real-time force feedback provided by force sensors mounted on the robot's fingertips. Overall, our solution combines DMPs, VSF and force feedback to achieve effective in-hand manipulation of unknown objects: while DMPs permit to extract the desired object trajectories from a single human demonstration, and to easily generalize them to new conditions, the VSF with force feedback allows to keep coordination between the fingers and proper contact forces with the object, without requiring an explicit model of the manipulated object (Fig.…”