Rigidity-Based Multi-Finger Coordination for Precise In-Hand Manipulation of Force-Sensitive Objects

Xinan Rong, Changhuang Wan, Aochen He, Xiaolong Li, Gangshan Jing

TL;DR

This paper introduces a novel dual-layer framework for precise in-hand manipulation of delicate objects using multi-fingered robotic hands. It achieves this by planning contact forces based on graph rigidity and then converting these forces into joint trajectories without relying on tactile sensors or torque control, demonstrating safe and accurate manipulation of fragile items like raw eggs.

What Problem Does This Paper Address?

Manipulating delicate objects with robotic hands is challenging because it requires precise control over both the object's position and the forces applied by the fingers. Unlike industrial grippers, multi-fingered hands rely on point contacts, making it harder to apply controlled forces. Furthermore, many commercial robotic hands lack the necessary torque sensors for fine force control, and tactile sensors, while helpful, are often expensive and complex. This paper tackles the problem of achieving high-precision in-hand manipulation of force-sensitive objects, ensuring stability and preventing damage, without relying on these common but often unavailable or impractical sensing and control modalities.

Methods

The core of the proposed solution is a dual-layer framework:

  1. Rigidity-Based Force Planner (RFP): This layer plans the necessary contact forces. It leverages a "Rigid Contact Framework" which models the hand-object interaction as a graph where vertices are contact points and edges represent distance constraints. This framework ensures the geometric configuration of the contact points remains invariant, implicitly regulating internal forces. The contact force is then decomposed into three components:
  2. Operational Force (fope): Balances external wrenches (like gravity).
  3. Rigidity Internal Force (fint,R): Enforces the rigid contact framework.
  4. Friction Internal Force (fint,µ): A supplementary force optimized to satisfy friction cone and safety limits, introduced only when needed. This decomposition simplifies the complex force planning problem into more tractable sub-problems.

  5. Force-to-Position Mapping and Control (FJC): This layer translates the planned contact forces into joint control commands. Since direct torque control is often unavailable, it uses a "compliance-based virtual penetration model." For each fingertip, a virtual target point is defined inside the object. By commanding the fingertip to track this virtual point using position control, the desired contact force is generated as the object's surface resists the motion. This force is then mapped to a penetration distance, which in turn defines the virtual target point. This process is integrated into a Model Predictive Control (MPC) loop for trajectory optimization and execution.

Datasets

Not discussed in the paper. The experiments were conducted on a custom robotic hand and used real-world objects.

Key Insights and Ideas

  • Rigidity as a Constraint for Internal Forces: The paper elegantly uses the concept of graph rigidity to implicitly control internal forces in multi-finger grasps. This is a departure from some approaches that try to suppress internal forces, recognizing their necessity for stable manipulation with many fingers.
  • Force Control via Virtual Penetration: By reformulating force control as a position tracking problem using a virtual penetration model, the authors bypass the need for direct torque control or tactile sensors, making the approach more practical for existing robotic hands.
  • Decoupled Force Decomposition: Breaking down the contact force into operational, rigidity, and friction components simplifies the planning problem and allows for targeted optimization of each aspect, leading to more robust and interpretable control.
  • Scalability: The graph rigidity approach is presented as scalable with the number of fingers, addressing a limitation in some existing multi-arm coordination methods.

Key Results

The framework was validated on a custom dexterous hand and demonstrated successful manipulation of:

  • Soft Yarn: Maintained tautness without stretching, confirming the rigid contact model.
  • Plastic Cup (empty and water-filled): Achieved stable grasps and accurate trajectory tracking, even with increased mass and challenging trajectories.
  • Raw Egg: Executed high-precision tracing of letters ("RAL") with a maximum waypoint error of only 0.3 mm, showcasing excellent force regulation and precision on a very fragile object.
  • Robustness: The system showed robustness to changes in gravity direction and parameter uncertainties (mass and friction coefficient), maintaining a safety margin.
  • Efficiency: Execution times per waypoint were competitive with existing methods for rigid object manipulation.

Limitations and Future Work

  • Object Properties: The compliance gains (Cx, Cy, Cz) for the force-to-position mapping need to be empirically tuned based on the target object's physical properties.
  • Model Accuracy: While the paper simplifies the dynamics model by neglecting inertial terms for lightweight objects, this might be a limitation for heavier objects.
  • Future Work: The authors suggest using learning-based models to automatically map contact forces and penetration, and employing online system identification to estimate object parameters. This would reduce manual tuning and prior dependence on physical properties.

Summary updated

Abstract

Precise in-hand manipulation of force-sensitive objects typically requires judicious coordinated force planning as well as accurate contact force feedback and control. Unlike multi-arm platforms with gripper end effectors, multi-fingered hands rely solely on fingertip point contacts and are not able to apply pull forces, therefore poses a more challenging problem. Furthermore, calibrated torque sensors are lacking in most commercial dexterous hands, adding to the difficulty. To address these challenges, we propose a dual-layer framework for multi-finger coordination, enabling high-precision manipulation of force-sensitive objects through joint control without tactile feedback. This approach solves coordinated contact force planning by incorporating graph rigidity and force closure constraints. By employing a force-to-position mapping, the planned force trajectory is converted to a joint trajectory. We validate the framework on a custom dexterous hand, demonstrating the capability to manipulate fragile objects—including a soft yarn, a plastic cup, and a raw egg—with high precision and safety.
BibTeX
@article{rong2026rigidity,
  author = {Xinan Rong and Changhuang Wan and Aochen He and Xiaolong Li and Gangshan Jing},
  title = {{Rigidity-Based Multi-Finger Coordination for Precise In-Hand Manipulation of Force-Sensitive Objects}},
  journal = {IEEE Robotics and Automation Letters},
  year = {2026},
  doi = {10.1109/lra.2026.3666396},
}
Figure