Minimal Work: A Grasp Quality Metric for Deformable Hollow Objects

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

TL;DR

Researchers have developed a new grasp quality metric called Minimal Work specifically for deformable hollow objects like plastic bottles and cups. By calculating the mechanical work required to resist external forces while minimizing object deformation, this approach allows robots to pick up fragile items securely without crushing them or spilling their contents.

What Problem Does This Paper Address?

In the world of robotics, we are quite good at grasping rigid objects. We use metrics like "force closure" to ensure a robot won't drop a brick. However, these traditional metrics fail when the object is "squishy."

If a robot treats a thin plastic cup like a piece of steel, it might apply enough force to prevent a slip but end up crushing the cup or splashing the coffee inside. Conversely, if it's too gentle, the cup falls. The challenge is finding the "Goldilocks" zone: applying just enough force to resist gravity and motion (the wrench) while minimizing the deformation of the object. Existing methods often rely on computationally expensive Finite Element Method (FEM) simulations, which are too slow for real-time robot planning.

Methods

The authors propose a metric based on the physical definition of work ($Work = Force \times Displacement$). Their pipeline consists of three main steps:

  1. Wrench Resistance: They use a linear program to find the minimum grasp force ($F$) required to resist a target external wrench (like gravity during a lift or rotation). They employ the REACH model to estimate contact area and a 6D ellipsoidal limit surface to model friction without needing a full simulation.
  2. Stiffness Mapping: Since the "squishiness" of a bottle varies (the neck is usually stiffer than the middle), they physically probe objects to create a stiffness map ($s$).
  3. Work Calculation: Using Hooke’s Law, they estimate the jaw displacement ($d = F/s$). The total "Minimal Work" is the product of the required force and this displacement.

By minimizing this work value, the robot naturally chooses grasp points that are both mechanically advantageous and located on stiffer parts of the object. To handle real-world uncertainty, they calculate this metric over several perturbed versions of a grasp to ensure reliability.

Datasets

The researchers didn't rely on a pre-existing dataset; they built a custom physical benchmark:

  • Objects: Five distinct objects, including off-the-shelf plastic detergent bottles and cups, as well as custom 3D-printed shapes made from NinjaFlex TPU (a flexible filament).
  • Physical Trials: They collected 460 physical grasps using a UR5 robot arm equipped with a Robotiq 2F-85 parallel-jaw gripper.
  • Success Criteria: A grasp was "successful" only if the robot completed the task (lifting or rotating) and the object returned to its original shape without spilling. They used a clever "balloon inflation" test inside the bottles to detect if the squeeze was hard enough to displace air/liquid.

Key Insights and Ideas

  • Decoupling is Key: The authors decoupled the wrench analysis from the deformation analysis. This allows them to avoid the "computational nightmare" of FEM simulations while still getting a high-fidelity estimate of how the object will behave.
  • Work as a Unified Metric: Using "Work" is an elegant way to combine two competing goals. High force is bad (crushing), and high displacement is bad (instability). Work penalizes both.
  • Geometry Matters: The paper highlights that for hollow objects, deformability depends more on local geometry and wall thickness than just the material. A robot that knows a bottle's "stiffness map" is significantly more capable than one that treats the whole object as a uniform mass.

Key Results

The "Minimal Work" (MW) metric significantly outperformed traditional rigid-body metrics:

  • Balanced Accuracy: The MW metric achieved 74.2% balanced accuracy in predicting grasp success.
  • Comparison: This was 24.2% higher than the standard Grasp Reliability metric and 12.7% higher than a metric that only considers minimal force.
  • Task Versatility: The metric proved robust for both simple vertical lifting and more complex 90-degree rotation tasks, where shifting gravitational loads often cause simpler grasping strategies to fail.

Limitations and Future Work

While a major step forward, the authors note a few hurdles:

  • Large Deformations: The linear elastic model ($F=ks$) is an approximation. For extremely soft objects that undergo massive shape changes, this linear assumption breaks down, leading to false positives.
  • Hardware Constraints: The Robotiq gripper used in the study sometimes applied more force than the "planned" minimum due to its internal controller, causing unintended deformations.
  • Future Directions: The team plans to test the metric on a wider variety of grippers (like the Schunk gripper) and explore the "duality" of the metric—applying it to rigid objects held by soft, deformable gripper pads.

Summary updated

Abstract

Robot grasping of deformable hollow objects such as plastic bottles and cups is challenging, as the grasp should resist disturbances while minimally deforming the object so as not to damage it or dislodge liquids. We propose minimal work as a novel grasp quality metric that combines wrench resistance and object deformation. We introduce an efficient algorithm to compute the work required to resist an external wrench for a manipulation task by solving a linear program. The algorithm first computes the minimum required grasp force and an estimation of the gripper jaw displacements based on the object's empirical stiffness at different locations. The work done by the jaws is the product of the grasp force and the displacements. Grasps requiring minimal work are considered to be of high quality. We collect 460 physical grasps with a UR5 robot and a Robotiq gripper. We consider a grasp to be successful if it completes the task without damaging the object or dislodging the content. Physical experiments suggest that the minimal work quality metric reaches 74.2% balanced accuracy, a metric that is the raw accuracy normalized by the number of successful and failed real-world grasps, and is up to 24.2% higher than classical wrench-based quality metrics.
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