Data for: A Machine Learning–Based Tool for Enhancing Position Accuracy in Industrial Robots with a Reduced Dataset

Here are shared the files related to the study presented in the paper:Title: "A Machine Learning–Based Tool for Enhancing Position Accuracy in Industrial Robots with a Reduced Dataset”
Authors: Giuseppe Romano, Pietro Bilancia, Alberto Locatelli, Mirko Mucciarini, Manuel Iori, Marcello Pellicciari

The repository is organized into three main folders:
01_Dataset
02_ML_Models
03_Compensation_Example

01_Dataset
This folder contains the experimental datasets used in the study.
• Preliminary_Sensitivity_Analysis_Dataset
Contains the datasets used for the preliminary analyses described in the paper.
These datasets investigate the influence of: Velocity variation (10%, 25%, 50%), Payload variation (39 kg, 79 kg, 105 kg), Approach direction and Grid granularity (27 points, 64 points, 125 points, 216 points).
These datasets were used to evaluate the sensitivity of positioning error to operating conditions before defining the reduced training configuration.
• Training_Dataset
Contains the datasets effectively used for training the machine learning models presented in the paper (each CSV file corresponds to a specific payload configuration).

02_ML_Models
This folder contains the trained machine learning models in .pkl format.
The following regressors are provided: CatBoostRegressor, GradientBoostingRegressor, LinearRegression, RandomForestRegressor, SGDRegressor, TabPFNRegressor and XGBRegressor.
These models were trained using the datasets provided in the Training_Dataset folder and are ready for direct inference.

03_Compensation_Example
This folder contains a practical example demonstrating how to apply the trained models for offline compensation.
It includes:
• compensation_script.py  Python script implementing the compensation procedure
• requirements.txt  Python dependencies with tested versions
• TabPFNRegressor.pkl  Example trained model
• test_10pt.dat  Robot program data file (input for compensation)
• test_10pt.src  Associated robot source file (not modified by the script)
The compensation procedure operates exclusively on the .dat file.