Designing Pin-pression Gripper and Learning its Dexterous Grasping with Online In-hand Adjustment

TL;DR

This work introduces a geometry-aware pin-pression gripper in which each finger contains a 4×4 array of independently actuated pins, enabling online shaping to diverse object geometries and in-hand re-orientation. A Soft Actor-Critic policy is trained with a two-stage curriculum to master both grasp-then-lift and grasp-while-lift modes, using a rich state representation and dual rewards to optimize success and efficiency. Extensive simulation and real-world experiments demonstrate robust generalization to unseen objects, superior performance over fixed-shape and passive baselines, and successful sim-to-real transfer via a teacher-student framework. The approach highlights the value of integrating adaptive hardware design with curriculum-driven learning for dexterous manipulation across a broad class of objects and poses, with practical implications for automated grasping in real-world settings.

Abstract

We introduce a novel design of parallel-jaw grippers drawing inspiration from pin-pression toys. The proposed pin-pression gripper features a distinctive mechanism in which each finger integrates a 2D array of pins capable of independent extension and retraction. This unique design allows the gripper to instantaneously customize its finger's shape to conform to the object being grasped by dynamically adjusting the extension/retraction of the pins. In addition, the gripper excels in in-hand re-orientation of objects for enhanced grasping stability again via dynamically adjusting the pins. To learn the dynamic grasping skills of pin-pression grippers, we devise a dedicated reinforcement learning algorithm with careful designs of state representation and reward shaping. To achieve a more efficient grasp-while-lift grasping mode, we propose a curriculum learning scheme. Extensive evaluations demonstrate that our design, together with the learned skills, leads to highly flexible and robust grasping with much stronger generality to unseen objects than alternatives. We also highlight encouraging physical results of sim-to-real transfer on a physically manufactured pin-pression gripper, demonstrating the practical significance of our novel gripper design and grasping skill. Demonstration videos for this paper are available at https://github.com/siggraph-pin-pression-gripper/pin-pression-gripper-video.

Figures (13)

• Figure 1: A picture of a pin-pression toy. The toy features a 2D array of pins capable of independent extension and retraction. When pressing over the pins, an imprint of the letter is formed, which is naturally adaptive to the letter’s shape.
• Figure 2: The parallel-jaw gripper (a) finds difficulty in grasping the trapezoid. Our pin-pression gripper (b and c) achieves shape-adaptive grasping with dynamic adjustment of pins. In the grasp-then-lift (GtL) mode (b), the gripper can form a better closure of the trapezoid by partially lifting it. The grasp-while-lift (GwL) mode (c) allows for dynamic adjustment of pins throughout the process of grasping and lifting and therefore has the opportunity to more tightly lock the trapezoid via in-hand re-orientation.
• Figure 3: Overview of our learning-based approach. Our method obtains the current state information about the object, gripper, and their interaction to predict the appropriate action that moves the pins of the gripper finger in GtL or GwL grasping modes. After executing the online predicted action, the updated state is then passed through the same pipeline to predict the next action, so that a successful grasp is gradually formed.
• Figure 4: The specific constitution of the parallel pin-pression gripper. The gripper finger is a 2D array of pins capable of independent extension and retraction. Each pin consists of a cylinder and a sphere. The gripper achieves full closure when every finger reaches its maximum forward movement.
• Figure 5: Informative dynamic interaction representations of each pin $p$, including a set of relationship information between the target object and the gripper. Detailed explanations of the notations can be found in Section \ref{['sec:state_action_representation']}.