cs.ROSep 24, 2026

A Support-Enhanced Granular-Jamming Gripper for RL-based Grasping with Continuum Manipulators

Authors: Danyu Liu, Tianlin Zhang, Wei Chen, Wei Tang, Kecheng Qin, Zhongyu Li

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Continuum manipulators provide dexterous motion in confined spaces, but structural compliance, hysteresis, and load-dependent deformation leave residual position and orientation errors that can undermine reliable contact with rigid grippers. To address this limitation, this paper presents a lightweight support-enhanced granular-jamming gripper tailored to a continuum manipulator. The gripper maintains compliance before jamming while establishing a direct load path to the continuum manipulator tip after jamming. To improve its grasping performance, we systematically designed membrane materials, particles, filling ratios, and the internal support structure, and further identify geometry-dependent grasp boundaries with respect to contact offset and object shape. Building on these results, we construct a physical manipulation system integrating the continuum manipulator, granular-jamming gripper, visual feedback, tendon actuation, and pneumatic control. We then train a reinforcement-learning-based reaching controller in a randomized simulation and deploy it on the physical system, demonstrating how positioning control and contact level mechanical adaptation can complement each other in a modular grasp-and-release task. By introducing an adaptive structure that relaxes the need for highly accurate modeling and positioning control, this work explores a design paradigm that integrates physical and embodied intelligence.

Figures & tables

Fig. 1: Motivation for mechanically adaptive grasping with a continuum manipulator. (A) A rigid gripper requires sufficiently accurate relative position, orientation, and geometric compatibility; residual deviations can cause collision or prevent stable closure. (B) The unjammed granular gripper deforms through contact and continued compression to envelop the object, after which vacuum-induced jamming locks the adapted configuration. The comparison illustrates how contact-induced conformity relaxes the requirement for a single precise pre-contact pose.
Fig. 2: Architecture of the autonomous grasping system. Visual feedback provides the target and gripper states to the reaching policy, which outputs a four-dimensional absolute tendon-length command. The task layer coordinates pneumatic jamming and release, while the gripper forms and retains contact through mechanical deformation. This functional separation enables autonomous operation without requiring the learned policy to model granular contact mechanics.
Fig. 3: Autonomous task sequence comprising policy-controlled approach, continued contact and enclosure, vacuum-induced jamming, lifting, transfer, and release. Only continuum reaching is learned; enclosure develops through passive mechanical deformation, while pneumatic switching is coordinated by the task layer. The sequence shows how learned reaching and mechanically adaptive grasp formation jointly produce the complete autonomous operation.
Fig. 4: Structure of the support-enhanced granular-jamming gripper. The flexible membrane and granular filler provide contact conformity, while the internal support promotes lateral enclosure, transmits compressive force during contact, and provides a direct load path after jamming. The structure therefore combines a compliant contact surface with internal force transmission suitable for a bending continuum tip.
Fig. 5: Operating sequence of the granular-jamming gripper. Contact and continued compression first redistribute the particles around the object, after which evacuation locks the adapted configuration. The pressure transition converts contact-generated conformity into a load-bearing grasp supported by the jammed particles and internal rod.
Fig. 6: Configuration selection and support-rod evaluation. (A) Grasp success rates for combinations of natural-rubber or thermoplastic-polyurethane membranes with expanded-polypropylene, polystyrene, or thermoplastic-rubber particles. (B) Success rates at nominal filling ratios of 1/31/3 , 1/21/2 , and 4/54/5 using the natural-rubber membrane and TPR spheres. (C) The selected configuration succeeds in 10/10 trials with the support rod and 6/10 trials without it. Each condition comprises ten trials. These results select the natural-rubber membrane, TPR spheres, and 50% filling ratio, and demonstrate that the internal support increases the measured grasp success rate from 60% to 100% under the tested condition.
Fig. 7: Geometry-dependent grasping under contact-position offset. (A) Each marked position is tested ten times, and the continuous field is interpolated from the discrete measurements only to visualize the spatial trend. (B) The membrane, support rod, object, and table jointly form a stable enclosure under favorable contact configurations. The shifted high-success region shows that the gripper enlarges a geometry-dependent set of graspable configurations rather than providing a uniform, isotropic position-tolerance region.
Fig. 8: Grasping performance across object geometries. (A) Representative sequences for a cylinder, cube, sphere, hexagonal prism, and triangular pyramid. (B) Their success rates are 80%, 80%, 100%, 90%, and 10%, respectively, over ten trials per shape. The results demonstrate adaptation to several geometries while identifying sharp-edged objects with limited stable contact as a limitation of the current design.
Fig. 9: Randomized training and zero-shot physical deployment of the reaching policy. In a task frame fixed to the continuum-manipulator base, the policy receives four consecutive observations of the gripper position, relative target displacement, and tendon state, and outputs a four-dimensional absolute tendon-length command. Randomizing the initial posture, target position, stiffness, and actuation delay exposes the policy to continuum-specific variations. The formulation preserves consistent state and action meanings between simulation and the physical system without requiring the policy to learn contact deformation or pneumatic control.
Fig. 10: Autonomous grasp-and-release sequence on the physical system: policy-controlled approach, contact and passive conformity, task-level vacuum activation, lifting, transfer, and venting over the destination. The completed sequence demonstrates that zero-shot continuum reaching, the support-enhanced gripper, and pneumatic task control can operate together without requiring the policy to model or control granular contact deformation.

Explore similar work

CardsList
Sep 21, 2026cs.RO

Steerable and Reactive Grasping Through Modular Design with a Three-Point Interface

Dexterous grasping requires deciding where to grasp, reaching the target, and maintaining stable contact. We connect these stages through a compact three-point interface that separates global geometric reasoning from local contact control. Given object geometry and optional language commands, our framework samples contact triples from a precomputed grasp-affordance heatmap. A model-based reactive controller tracks the object, avoids collisions, and guides the hand toward the selected contacts. In the final centimeters, a Reinforcement Learning (RL) policy uses proprioceptive feedback to refine and stabilize the grasp despite reaching and perception errors. It observes only finger joint states and its recent actions, with no target points, visual observations, or object geometry, so a single policy is shared across objects and grasp configurations. In simulation, we compare grasp-and-lift success against squeeze and end-to-end baselines, characterize reaching convergence, and demonstrate grasp steering; hardware demonstrations on two training objects and one unseen object illustrate the full pipeline. Our modular framework uses geometry to guide the reach and local feedback to secure the grasp.

Andrew Nguyen, Yonghyeon Lee, Sangbae Kim

University of Michigan, Ann Arbor, MI, USA. · Department of Mechanical Engineering, MIT, Cambridge, MA, USA. · Department of AI, Yonsei University, Seoul, South Korea.

Robotic GraspingProprioceptionModular FrameworkSteering
AskPDF
Jun 26, 2026cs.RO

Learning Stable In-Grasp Manipulation in a Non-Dropping Action Space

Traditionally, dexterous manipulation controllers are designed using analytic models constrained by strong assumptions about the hand and the objects being manipulated. Reinforcement learning (RL) has become another common approach in which skills are explored openly in an end-to-end manner but is inefficient because of unnoticeable instability and conflicts in learning objectives. This paper attempts to efficiently explore stable and accurate manipulation skills by decomposing dexterous skills into multiple simpler/analyzable components. Each skill component is subsequently learned with constraints and guidance from classical physics and control theory. Our work shows that for stable grasp, in-grasp reposition/reorientation with different objects, sensor/motor noise, latency, and frictional conditions, skill learning becomes efficient and stable with prior knowledge from theory.

Ha Thang Long Doan, Hikaru Arita, Kazuto Nakashima +1

Robotic GraspingDexterous ManipulationAction Space
AskPDF
Apr 14, 2026cs.RO

FastGrasp: Learning-based Whole-Body Control Method for Fast Dexterous Grasping with Mobile Manipulators

Fast dexterous grasping while a mobile base remains in motion requires coordinated whole-body control and rapid adaptation to physical contact. We propose FastGrasp, a two-stage learning framework that integrates grasp guidance, whole-body control, and tactile feedback. First, a pretrained conditional variational autoencoder generates diverse grasp candidates from object point clouds. Candidates are filtered by approach direction and supporting-surface constraints, then ranked using an envelopment-based criterion combining grasp width and depth coverage measures. Second, a reinforcement learning policy jointly controls the mobile base, arm, and dexterous hand, using the selected grasp as guidance and tactile observations for online grasp adjustment. The policy is trained with domain randomization and deployed with command filtering and tactile-triggered grasp tightening. Simulation experiments show higher grasp success rates than the evaluated baselines under full and partial point-cloud observations, while real-world experiments demonstrate sim-to-real transfer across diverse object geometries.

Heng Tao, Yiming Zhong, Zemin Yang +1

ShanghaiTech University

Robotic GraspingWhole-Body Control
AskPDF