[Paper Review] Dense Tactile Force Distribution Estimation using GelSlim and inverse FEM

Daolin Ma, Elliott Donlon+2 more|arXiv (Cornell University)|Oct 10, 2018

Advanced Sensor and Energy Harvesting Materials21 references19 citations

TL;DR

This paper presents GelSlim 2.0, a vision-based tactile sensor that estimates dense 3D force distributions in real time using inverse Finite Element Method (iFEM) to reconstruct forces from marker displacement data on a deformable gel pad. The method achieves high accuracy, with reconstructed resultant forces within 15% of ground truth from a six-axis force-torque sensor, demonstrating physically plausible and spatially consistent force fields.

ABSTRACT

In this paper, we present a new version of tactile sensor GelSlim 2.0 with the capability to estimate the contact force distribution in real time. The sensor is vision-based and uses an array of markers to track deformations on a gel pad due to contact. A new hardware design makes the sensor more rugged, parametrically adjustable and improves illumination. Leveraging the sensor's increased functionality, we propose to use inverse Finite Element Method (iFEM), a numerical method to reconstruct the contact force distribution based on marker displacements. The sensor is able to provide force distribution of contact with high spatial density. Experiments and comparison with ground truth show that the reconstructed force distribution is physically reasonable with good accuracy.

Motivation & Objective

• To develop a vision-based tactile sensor capable of real-time dense force distribution estimation, overcoming limitations of prior methods that focus only on geometry or assume local force-displacement proportionality.
• To address the physical inaccuracy of force reconstruction when assuming force is proportional only to local displacement, which fails to account for internal stress propagation in elastic materials.
• To improve tactile sensing hardware by creating a more rugged, parametrically adjustable, and better-illuminated version of GelSlim with enhanced marker tracking fidelity.
• To validate the proposed inverse FEM (iFEM) method for reconstructing global force distributions from measured deformation fields, ensuring consistency with continuum mechanics principles.
• To demonstrate that the sensor can produce physically realistic force fields that match ground truth measurements from a six-axis force-torque sensor with minimal calibration error.

Proposed method

• The sensor uses a vision-based system with an array of printed markers on a silicone gel pad to track surface deformations under contact via high-resolution imaging.
• A new hardware design improves mechanical robustness, illumination uniformity, and parametric adjustability for consistent marker tracking across different contact conditions.
• The core method is inverse Finite Element Method (iFEM), which solves for the external force distribution by inverting the elastostatic relationship between measured displacements and applied forces.
• iFEM formulates a global weak form of the equilibrium equation, using the measured displacement field as input and solving for the external force field that would produce such deformation under known material and geometric parameters.
• The method accounts for the full spatial coupling of forces and displacements, avoiding the oversimplified assumption that force at a point depends only on its own displacement.
• Normal force distribution is estimated by analyzing the 3D geometry of the contact patch via image processing and known sphere radius, enabling consistent normal force reconstruction.

Experimental results

Research questions

• RQ1 Can a vision-based tactile sensor with high spatial resolution and real-time processing estimate dense 3D force distributions with physical plausibility?
• RQ2 Does inverse FEM (iFEM) outperform local force-displacement models in reconstructing force fields from marker displacement data, especially in capturing non-local stress effects?
• RQ3 To what extent does the reconstructed force distribution match ground truth measured by a high-accuracy six-axis force-torque sensor?
• RQ4 How does the new hardware design of GelSlim 2.0 improve measurement fidelity and robustness compared to previous versions?
• RQ5 Can the method reliably reconstruct both tangential and normal force components with consistent spatial patterns across diverse contact geometries?

Key findings

• The reconstructed force distribution using iFEM is physically plausible, with tangential forces naturally confined to the contact patch and no unphysical force spikes outside the contact area.
• The standard deviation of the reconstructed resultant force across x, y, and z directions was (0.244N, 0.201N, 0.322N), representing approximately 15% error relative to ground truth from a six-axis force-torque sensor.
• The comparison between reconstructed and ground-truth forces shows strong alignment along the identity line, indicating high accuracy and low overfitting, with results dependent only on independently measured geometry and material parameters.
• The normal force distribution for spherical contact is smoothly distributed and consistent with theoretical expectations, confirming the validity of the normal displacement estimation method.
• The sensor successfully captures complex manipulation dynamics, such as in Kendama tasks, with force fields that align with critical phase transitions like grasping and releasing.
• The method demonstrates robustness to noise and non-uniform deformations, including cases where stretched fabric causes localized pressure not fully captured in normal displacement, yet the force reconstruction remains stable and realistic.