Touch with Insight: Physics-Aware Data-Driven Learning for EIT-Based Tactile Sensing
Kiyanoush Nazari, Yunqi Huang, David Hardman, Fumiya Iida, Thomas George Thuruthel
AI summary
Problem
Accurately estimating contact location and force from Electrical Impedance Tomography (EIT) sensors is hindered by an ill-posed, noisy, and highly nonlinear inverse mapping that pure data-driven models struggle to generalize across.
Approach
The authors combine a neural predictor with a physics-informed latent regularizer that enforces consistency between predicted tactile states and a voltage manifold learned via an autoencoder.
Key results
- Physics-informed latent regularization consistently improves force estimation accuracy over pure data-driven baselines
- Comprehensive benchmark across MLPs, CNNs, Transformers, and autoencoder regressors establishes clear performance baselines
- High-fidelity simulation pipeline incorporating sim-to-real gap factors enables robust synthetic data generation
- Hybrid framework demonstrates strong cross-domain generalization on both real and synthetic datasets
Why it matters
Provides a scalable, physics-grounded framework for reliable tactile state estimation, advancing dexterous robotic manipulation and sensor design.
Abstract
Tactile sensing is essential for enabling dexterous robotic manipulation, yet estimating contact states such as location and force from high-dimensional sensor measurements remains challenging due to noise and complex nonlinear map- pings between raw signals and physical interaction states. In this work, we propose a physics-informed contact modeling framework that combines the flexibility of deep models with inductive biases from physical modeling. Focusing on electrical impedance tomography (EIT) tactile skins, our approach incor- porates knowledge of the EIT forward model by regularizing neural estimators with a latent-space consistency constraint, stabilizing the ill-posed inverse mapping from voltages to contact states. To support robust training and evaluation, we also develop a high-fidelity simulation pipeline that incorpo- rates key hardware imperfections to better bridge the sim- to-real gap. We benchmark multiple architectures—including multilayer perceptrons, convolutional networks, Transformer- based models, and autoencoder regressors—on both real and synthetic datasets. Results show that the proposed hybrid ap- proach consistently improves estimation accuracy, particularly for force prediction, and generalizes across domains. These findings highlight the value of embedding physical priors into learning pipelines for reliable tactile state estimation in robotic manipulation. (video, code)