Minimal Work: A Grasp Quality Metric for Deformable Hollow Objects

Jingyi Xu, Michael Danielczuk, Jeffrey Ichnowski, Jeffrey Mahler, Eckehard G. Steinbach, Ken Goldberg

Introduction

Robotic manipulation has achieved significant success with rigid objects using metrics such as force closure. However, these traditional approaches often fail when applied to deformable hollow objects (DHOs), such as plastic bottles or thin-walled cups. Treating a DHO as a rigid body can lead to excessive force that crushes the object or insufficient force that results in slippage. The primary challenge lies in identifying a "Goldilocks" zone: applying enough force to resist external wrenches (e.g., gravity) while minimizing structural deformation.

Methodology

The researchers propose a novel metric, Minimal Work (MW), based on the physical principle that work equals force times displacement ($W = F \cdot d$). The approach avoids computationally intensive Finite Element Method (FEM) simulations by decoupling wrench analysis from deformation modeling.

The Three-Step Pipeline

  1. Wrench Resistance: The system calculates the minimum grasp force ($F$) required to resist a target external wrench. This is achieved using a linear program that incorporates the REACH model for contact area estimation and a 6D ellipsoidal limit surface to model friction.
  2. Stiffness Mapping: Recognizing that DHOs exhibit non-uniform compliance (e.g., a bottle neck is stiffer than its center), the researchers generate a stiffness map ($s$) through physical probing.
  3. Work Calculation: Applying Hooke’s Law ($F = s \cdot d$), the jaw displacement is estimated as $d = F/s$. The Minimal Work metric is then defined as the product of the required force and this estimated displacement ($W = F^2/s$).
Stiffness mapping and grasp planning for a deformable plastic cup.

By minimizing this work value, the robot naturally prioritizes grasp points that are mechanically efficient and located on stiffer regions of the object.

Experimental Evaluation

To validate the MW metric, a custom physical benchmark was established featuring five distinct objects: commercial detergent bottles, plastic cups, and 3D-printed NinjaFlex TPU shapes. The researchers conducted 460 physical grasp trials using a UR5 robot arm and a Robotiq 2F-85 parallel-jaw gripper. Success was defined by the robot's ability to lift or rotate the object without permanent deformation or spilling contents, verified via an internal "balloon inflation" pressure test.

Key Results

The Minimal Work metric demonstrated superior performance compared to traditional rigid-body and force-only metrics.

  • Predictive Accuracy: MW achieved a 74.2% balanced accuracy in predicting grasp success.
  • Comparative Gains: This represents a 24.2% improvement over the standard Grasp Reliability metric and a 12.7% improvement over metrics considering only minimal force.
  • Task Robustness: The metric remained effective across both simple vertical lifts and complex 90-degree rotations, where shifting gravitational loads typically challenge simpler strategies.
Comparative performance of grasp metrics across various objects and tasks.

Discussion and Future Work

While the MW metric offers a significant advancement, it relies on a linear elastic approximation ($F=ks$). This assumption may break down for extremely soft objects undergoing large non-linear deformations. Additionally, hardware limitations, such as the internal controller of the Robotiq gripper, occasionally resulted in higher-than-planned forces.

Future research will explore the application of the MW metric to a broader range of end-effectors and investigate the "duality" of the metric—applying it to rigid objects manipulated by soft, deformable gripper pads.

Summary updated

BibTeX
@inproceedings{xu2020minimal,
  author = {Jingyi Xu and Michael Danielczuk and Jeffrey Ichnowski and Jeffrey Mahler and Eckehard G. Steinbach and Ken Goldberg},
  title = {{Minimal Work: A Grasp Quality Metric for Deformable Hollow Objects}},
  booktitle = {IEEE International Conference on Robotics and Automation},
  year = {2020},
  doi = {10.1109/icra40945.2020.9197062},
}
Figure