Abstract:Estimating the location of contact is a primary function of artificial tactile sensing apparatuses that perceive the environment through touch. Existing contact localization methods use flat geometry and uniform sensor distributions as a simplifying assumption, limiting their ability to be used on 3D surfaces with variable density sensing arrays. This paper studies contact localization on an artificial skin embedded with mutual capacitance tactile sensors, arranged non-uniformly in an unknown distribution along a semi-conical 3D geometry. A fully connected neural network is trained to localize the touching points on the embedded tactile sensors. The studied online model achieves a localization error of $5.7 \pm 3.0$ mm. This research contributes a versatile tool and robust solution for contact localization that is ambiguous in shape and internal sensor distribution.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: how to accurately locate the contact point on a three - dimensional (3D) tactile artificial skin with non - uniform sensor distribution. Existing contact location methods usually assume that the sensors are uniformly distributed and the geometry is planar, which limits their application on complex three - dimensional surfaces. This paper proposes a machine - learning - based method to locate the contact points of the embedded mutual - capacitance tactile sensors by training a fully - connected neural network. ### Specific problem description: 1. **Limitations of existing methods**: - Existing contact location methods rely on flat geometric structures and uniformly distributed sensors, which makes them difficult to be applied to three - dimensional surfaces with complex shapes and non - uniform sensor distribution. - These methods usually assume that the sensor array is planar or the sensor density is uniform, thus limiting their flexibility and accuracy in practical applications. 2. **Research objectives**: - This paper aims to develop a method that can accurately locate contact points on a three - dimensional tactile artificial skin with non - uniform sensor distribution. - By using mutual - capacitance tactile sensors and fully - connected neural networks, the researchers hope to overcome the limitations of existing methods and provide a more general and robust solution. ### Solutions: - **Sensor design and arrangement**: The researchers designed a curved - surface, variable - density artificial skin, and the sensors are distributed in a non - uniform manner on a semi - conical geometric structure. - **Data collection and calibration**: The data of touch positions are collected by random sampling and equal - spacing sampling, and are used to train the neural network. - **Model training**: A fully - connected neural network is used to map the original sensor readings to the contact positions, minimizing the error between the predicted positions and the actual positions. - **Result evaluation**: The accuracy and uncertainty of the model are evaluated by comparing the known touch positions and the model - predicted positions. ### Main contributions: - A contact location method suitable for complex three - dimensional surfaces and non - uniform sensor distribution is proposed. - A positioning error of 5.7±3.0 millimeters is achieved, demonstrating the potential of this method in practical applications. - A flexible and robust tool is provided, which can handle cases where the shape and internal sensor distribution are uncertain. In summary, this paper solves the limitations of existing contact location methods in the application on complex three - dimensional surfaces, proposes a new machine - learning - based method, and significantly improves the accuracy and application range of contact location.

A Machine Learning Approach to Contact Localization in Variable Density Three-Dimensional Tactile Artificial Skin

A Novel Inverse Solution Of Contact Force Based On A Sparse Tactile Sensor Array

A Sensor Position Localization Method for Flexible, Non-Uniform Capacitive Tactile Sensor Arrays

Learning Decoupled Multi-touch Force Estimation, Localization and Stretch for Soft Capacitive E-skin

Non-contact Parametric Estimation and Localization of Human Head for Transcranial Magnetic Stimulation (TMS)

Discrimination Of Object Curvature Based On A Sparse Tactile Sensor Array

Ground truth force distribution for learning-based tactile sensing: a finite element approach

Large-area magnetic skin for multi-point and multi-scale tactile sensing with super-resolution

Real-Time Position Detecting of Large-Area CNT-based Tactile Sensors based on Artificial Intelligence

Machine Learning for Touch Localization on Ultrasonic Wave Touchscreen

Machine Learning for Touch Localization on an Ultrasonic Lamb Wave Touchscreen

Model-Based 3D Contact Geometry Perception for Visual Tactile Sensor

Sim-to-Real for High-Resolution Optical Tactile Sensing: From Images to Three-Dimensional Contact Force Distributions

Localizing the Object Contact through Matching Tactile Features with Visual Map

Tactile Super-Resolution Model for Soft Magnetic Skin

Learning Complex 3D Human Self-Contact

Multimodal Electronic Skin Integrated with 3-D Force Detection and High Precision Position Estimation.

3D Visual Tracking to Quantify Physical Contact Interactions in Human-to-Human Touch

Contact modelling and tactile data processing for robot skin

Design a Multifunctional Soft Tactile Sensor Enhanced by Machine Learning Approaches

3D force identification and prediction using deep learning based on a Gelsight-structured sensor