Optimizing BioTac Simulation for Realistic Tactile Perception

TL;DR

This paper investigates a BioTac simulation that uses temperature, force, and contact point positions to predict the sensor outputs, and reveals that the XGBoost regressor and transformer outperform traditional feed-forward neural networks in this task.

Abstract

Tactile sensing presents a promising opportunity for enhancing the interaction capabilities of today's robots. BioTac is a commonly used tactile sensor that enables robots to perceive and respond to physical tactile stimuli. However, the sensor's non-linearity poses challenges in simulating its behavior. In this paper, we first investigate a BioTac simulation that uses temperature, force, and contact point positions to predict the sensor outputs. We show that training with BioTac temperature readings does not yield accurate sensor output predictions during deployment. Consequently, we tested three alternative models, i.e., an XGBoost regressor, a neural network, and a transformer encoder. We train these models without temperature readings and provide a detailed investigation of the window size of the input vectors. We demonstrate that we achieve statistically significant improvements over the baseline network. Furthermore, our results reveal that the XGBoost regressor and transformer outperform traditional feed-forward neural networks in this task. We make all our code and results available online on https://github.com/wzaielamri/Optimizing_BioTac_Simulation.

Figures (9)

• Figure 1: (a): Map of the electrodes on the BioTac sensor Lin2013. (b): BioTac sensor and the position/ orientation of the coordinate frame Lin2013.
• Figure 2: Temperature values of the BioTac sensor in the Ruppel et al. Ruppel2018's dataset. The dashed black line represents the average temperature value.
• Figure 3: Average normalized MAE over all channels for ten folds for Network $B$. The solid line represents the normalized MAE values when fixing the temperature input value and probing all temperatures in the dataset. The dashed line represents the normalized MAE, when using the appropriate correct temperature values of the test set. The shaded area represents the upper and lower limits calculated by the standard deviation. The vertical dash-dotted line represents the mean temperature of the entire dataset.
• Figure 4: Visualization of the training and validation loss values for all investigated conditions, showing no sign of overfitting. The solid line represents the mean training loss function over all ten folds. The dashed lines depict the mean validation loss over all ten folds.
• Figure 5: Visualization of our used transformer architecture, based on Dosovitskiy et al. Dosovitskiy2021ViT. L stands for the number of attention blocks used in the transformer encoder.