Constructing Contact Estimation Models for Barometric Tactile Sensors
Sharmi Shah, Ethan Chun, Hongmin Kim, Andrew SaLoutos, David Nguyen, TaeWon Seo, Sangbae Kim
AI summary
Problem
Complex barometric tactile sensors suffer from time-dependent viscoelastic effects that degrade data-driven contact estimation accuracy. Prior inference methods struggle to handle these dynamics reliably on non-planar geometries while running on resource-constrained hardware.
Approach
The authors introduce a lightweight Binned-RNN architecture that discretizes contact angles for classification and incorporates temporal context to model elastomer dynamics, paired with a standardized data collection protocol featuring uniform surface sampling and explicit contact events.
Key results
- 30.4% reduction in normal force prediction error via improved data collection
- Binned-RNN eliminates edge localization bias and ensures uniform accuracy across curved surfaces
- Achieves 0.86 mm spatial resolution at 100 Hz inference on an integrated microcontroller
- Explicit contact flag detection outperforms force thresholding under material relaxation
Why it matters
Provides a scalable, hardware-efficient framework for deploying accurate tactile feedback on complex, low-cost robotic sensors.
Abstract
Barometric tactile sensors present a cheap and customizable method for adding tactile sensing to robotic platforms. These sensors consist of commercially available MEMS barometers embedded in an elastomer. However, as the sensing surface and elastomer volume increase in complexity, time-dependent material dynamics reduce sensing accuracy. We present a collection of inference and usage recommenda- tions towards mitigating these dynamics and improving sensor force and localization resolution. Using two custom, curved, barometric tactile sensors as case studies, we demonstrate that a new data collection regime alone can improve normal force predictions by 30.4% compared to prior work. We further introduce a Binned-RNN inference architecture and demonstrate its efficacy through select ablations. Small enough to run on the sensor’s integrated microcontroller at 100Hz, we find our model achieves a minimum spatial resolution of 0.86 mm on an ellipsoid tactile sensor. Finally, we demonstrate the robustness of these sensing capabilities through freeform contact and controlled object rolling. Video demonstrations can be found at https://youtu.be/mi7qqjssirg.