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

Description

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.

Steps to reproduce

To reproduce the compensation process described in the paper, follow the steps below. 1. Create a Python Virtual Environment It is recommended to create a clean environment to ensure compatibility with the provided models. 2. Install Required Dependencies Install the tested package versions: pip install -r requirements.txt 3. Run the Compensation Script Execute the script as follows: python compensation_script.py -i test_10pt.dat -m TabPFNRegressor.pkl -p 45 Where: • -i  Input robot .dat file containing the commanded Cartesian positions to be compensated • -m  Trained machine learning model (.pkl format) • -p  Payload value in kilograms 4. Output After execution, the script automatically creates a folder: /output/ Inside this folder, a new .dat file is generated containing the compensated Cartesian coordinates. The original .src file is not modified.

Institutions

• University of Modena and Reggio Emilia

  Emilia-Romagna, Modena

Categories

Licence