This repository contains the MATLAB code, experimental datasets, and simulation resources accompanying the article:

Goshtasbi A., Berghuis M., Parvaresh A., Murali Babu S. P., Style R.W., Rafsanjani A., Tactile Perception through Fluid–Solid Interaction, Nature Communications (2026).

Raw Data Processing

The Raw Data Processing folder includes MATLAB scripts used to process raw analog sensor signals and experimental force data to generate features and labels for machine learning.

The provided MATLAB scripts:

• Process raw analog signals from the sensor
• Extract input features from pressure signals
• Process experimental force measurements to generate labels

Processing pipeline

1. Remove baseline offset from raw signals
2. Extract from sensor signals
   • Maximum pressure (peak value)
   • Rise time (time to peak)
3. Extract from experimental data
   • Maximum force (used as label)

Example datasets are provided for:

• 1D preliminary results for all 3 media

• 2D preliminary results for oil-filled sensor

• Underwater 1D sensor

How to use

1. Add the folder to MATLAB path:

   addpath(genpath('Raw_Data_Processing'));

2. Load an example dataset loaded in preliminary or underewater folder:

   data = readmatrix('example_dataset.mat');

3. Run the processing script

   process_raw_data
   

Machine Learning

The Machine Learning folder contains MATLAB code used to map sensor signals to position and force using a combination of Fuzzy C-Means (FCM) clustering and Adaptive Neuro-Fuzzy Inference System (ANFIS). A detailed description of the model is provided in Supplementary Information S11 of the paper.

The MATLAB script:

• Learns the relationship between pressure values (input) and force and position (outputs)
• Predicts:
  • Position (X, X-Y, or Hilbert index)
  • Force (F)
• Supports both 1D and 2D sensor

Workflow

1. Select sensor type (1D or 2D)
2. Select dataset (provided in this folder)
3. Define the number of clusters (FCM)
4. Train FIS model using ANFIS
5. Evaluate model performance (regression + RMSE)
6. Save trained models

Included data

Example datasets are provided in this folder and correspond to those used in the paper:

• 1D sensor (all data)
• 2D sensor (60×60 sample)

How to run

3. Run the processing script

   addpath(genpath('Learning'));
   run('main_training_script.m')   % replace with your script name
   

Then:

• Select 1D or 2D sensor
• Choose one of the provided .xlsx dataset
• Enter the number of clusters when prompt

Output

• Trained FSI models (.mat and .fis files)
• Regression plots for each predicited variable
• Test RMSE displayed in MATLAB console

Model details

• Clustering: Fuzzy C-Means (via genfis)
• FIS type: Sugeno
• Training: ANFIS (hybrid optimization)
• Trian/test split: 80%/20% (randomized)
• Separated models are trained for each output (X, Y, F)

Notes

• Input structure:
  • 1D: PL, PR
  • 2D: PL, PR, TL, TR
• Output structure:
  • 1D: X, F
  • 2D: X, Y, F
• Results depend o the chosen number of clusters (please refer to the paper for better understanding the trade-off)
• Ensure dataset format matches the expected column structure

MATLAB–Python Transmission Control Protocol (TCP)

This folder enables real-time sensing and robot control by establishing a TCP/IP connection between MATLAB and Python. It integrates:

• MATLAB → signal processing + ANFIS prediction
• Python → robot control (UR robot via RTDE)

Goal

• Perform real-time tactile inference from sensor data
• Send predicted position (X) and force (F → Z motion) from MATLAB to Python
• Control a UR robot for human–robot interaction (HRI) experiments

Overview

1. MATLAB:
   • Acquires sensor data (NI-DAQ)
   • Detects contact events
   • Extracts features (PL, PR)
   • Predicts:
     • X position
     • Force (F)
   • Sends data (X,Z) via TCP
2. Python:
   • Receives TCP data (X,Z)
   • Converts predictions into robot motion
   • Executes poke motion using UR RTDE interface

How to run

1. Start Python server (Robot control) The code:
   • Initializes robot connection
   • Moves robot to initial position
   • Starts TCP server (127.0.0.1:65432)
2. Run MATLAB script The code:
   • Loads trained ANFIS models
   • Connects to Python via TCP
   • Starts real-time monitoring and predicition

Requirements MATLAB

• Data Acquisition Toolbox (NI-DAQ)
• Fuzzy Logic Toolbox Python
• Socket
• rtde_control
• rtde_receive

Notes

• MATLAB must start after Pythonserver is running
• TCP communications uses:
  • IP: 127.0.0.1
  • Port: 65432
• Force is mapped to Z displacement with safety clamping
• Robot motion includes: Approach → poke → return sequence

Fluid-Solid Interaction (FSI) COMSOL Simulations

This section contains files and instructions related to the Fluid–Solid Interaction (FSI) simulation performed in COMSOL Multiphysics. A detailed description and simulation data can be found at the following link