In-Hand Singulation and Scooping Manipulation with a 5 DOF Tactile Gripper

Yuhao Zhou, Pokuang Zhou, Shaoxiong Wang, Yu She

TL;DR

This paper introduces a novel, dexterous 5-DOF robotic gripper equipped with a GelSight tactile sensor. This gripper excels at challenging manipulation tasks, including singulating objects from granular media using only touch and performing precise scooping and insertion of flat objects like credit cards. The work demonstrates that a simpler, 5-DOF design can achieve high performance in complex manipulation scenarios, reducing the complexity often associated with multi-finger hands.

What Problem Does This Paper Address?

Robotic manipulation often requires high dexterity, traditionally achieved with complex, multi-finger hands with many degrees of freedom (DOF). While these hands offer versatility, their inherent redundancy can lead to complex control strategies and inefficient learning. This paper tackles the challenge of achieving dexterous manipulation with a simpler, more manageable gripper. Specifically, it addresses two key problems:

  1. Singulation from Granular Media: Extracting a single object from a pile of granular material (like rice) is difficult, especially when the object is partially buried. Existing methods often rely heavily on vision, which can be hindered by the granular medium. This paper aims to perform this task using primarily tactile feedback, including removing residual granular material adhering to the object.
  2. Precise Scooping and Insertion of Flat Objects: Grasping and manipulating thin, flat objects like credit cards, especially in confined spaces where robot arm movement is limited, is challenging. Traditional grippers may struggle with the precise orientation and insertion required for tasks like payment terminals. This paper seeks to enable such manipulation using in-hand dexterity.

Methods

The core of this work is the design and application of a novel 5-DOF tactile gripper.

  • Gripper Design: The gripper features two fingers, each with two degrees of freedom: one linear actuator for extension/retraction and one rotation servo for angular movement. This provides a total of 4 DOFs for the fingers, plus the inherent parallel motion of the base gripper. A key innovation is the integration of a GelSight tactile sensor on one fingertip, providing rich contact information. The other finger has a compliant dummy pad. The gripper's arms are 3D printed from PLA for lightweight construction.
  • Tactile-Reactive MPC for Grasping: For stable grasping in the granular media task, the authors employ a tactile-reactive Model Predictive Control (MPC) controller. This controller uses feedback from the GelSight sensor to dynamically adjust the grip, ensuring robustness and adaptability during manipulation. The cost function is specifically tailored to focus on contact area feedback.
  • Rubbing Motion with Linear Actuators: To address the challenge of removing granular media, a linear state feedback controller is proposed. This controller regulates the linear actuator's movement range, correlating it with the gripping width to achieve a rubbing motion that dislodges impurities.
  • Scooping Manipulation Maneuver: For handling flat objects like credit cards, a novel scooping maneuver is devised. This involves using a replaceable PLA "fingernail" on one finger to facilitate the initial scoop. The 5-DOF design allows for in-hand reorientation and precise adjustment of the card's position and orientation.
  • Force Analysis for Scooping: The paper includes a detailed force analysis using a simplified model to understand the mechanics of the scooping maneuver, particularly the moments and forces acting on the credit card.

Datasets

The paper does not explicitly mention training datasets. However, it uses a custom experimental setup for evaluation:

  • Singulation Task: Objects (small nuts, seeds, rigid balls, a soft foam sphere, an artificial strawberry, a golf ball) were embedded in rice within a plastic container.
  • Credit Card Insertion Task: A 3D-printed credit card and a custom-printed card reader were used.

The experiments involve multiple trials (e.g., 15 trials per object for singulation) to evaluate performance.

Key Insights and Ideas

  • Dexterity with Simplicity: The research strongly suggests that a 5-DOF gripper, augmented with tactile sensing, can achieve sophisticated manipulation tasks that might otherwise require much more complex robotic hands. This points towards a more efficient design paradigm for certain manipulation challenges.
  • Touch is Powerful for Singulation: The success in singulating objects from granular media using primarily tactile feedback highlights the potential of touch sensing for tasks where vision is occluded or unreliable. The ability to "feel" the object and its surroundings is crucial.
  • In-Hand Reorientation is Key: The paper demonstrates the effectiveness of in-hand manipulation for tasks like orienting a credit card. This ability to adjust an object's pose within the gripper, without moving the robot arm, is vital for precise insertion and assembly.
  • Tactile Feedback for Fine Control: The GelSight sensor plays a critical role, not just in detecting contact but in providing rich information that enables precise control for both rubbing off impurities and orienting flat objects.

Key Results

  • Singulation:
    • An overall success rate of 76% for the rubbing motion (singulation and impurity removal) across 10 different objects.
  • A remarkably high success rate of 94.3% for sphere-shaped objects.
  • Classification of singulated objects showed high accuracy, confirming the effectiveness of the tactile-based approach.
  • Credit Card Insertion:
    • Achieved a 100% success rate for the complete task (scooping and precise insertion) in 10 consecutive trials, even with random initial poses of the credit card.

Limitations and Future Work

  • Non-Spherical Object Singulation: The paper acknowledges that the success rate for non-spherical objects was lower. This is attributed to variability in width during rotation, making them less stable and prone to being dropped or squeezed out.
  • Learning-Based Approaches: The authors suggest that learning-based approaches could potentially improve performance for handling these more challenging, non-spherical objects.
  • Future Work: The authors aim to expand their approach to tackle even more intricate manipulation tasks, implying a desire to push the boundaries of what can be achieved with their gripper design.

Summary updated

Abstract

Manipulation tasks often require a high degree of dexterity, typically necessitating grippers with multiple degrees of freedom (DoF). While a robotic hand equipped with multiple fingers can execute precise and intricate manipulation tasks, the inherent redundancy stemming from its extensive DoF often adds unnecessary complexity. In this paper, we introduce the design of a tactile sensor-equipped gripper with two fingers and five DoF. We present a novel design integrating a GelSight tactile sensor, enhancing sensing capabilities and enabling finer control during specific manipulation tasks. To evaluate the gripper’s performance, we conduct experiments involving two challenging tasks: 1) retrieving, singularizing, and classification of various objects embedded in granular media, and 2) executing scooping manipulations of credit cards in confined environments to achieve precise insertion. Our results demonstrate the efficiency of the proposed approach, with a high success rate for singulation and classification tasks, particularly for spherical objects at high as 94.3%, and a 100% success rate for scooping and inserting credit cards.
BibTeX
@misc{zhou2024hand,
  author = {Yuhao Zhou and Pokuang Zhou and Shaoxiong Wang and Yu She},
  title = {{In-Hand Singulation and Scooping Manipulation with a 5 DOF Tactile Gripper}},
  year = {2024},
  eprint = {2408.00610},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2408.00610},
}
Figure