Human-Robot Collaboration (HRC) Simulation Artifacts

This repository contains the research artifacts for the paper "Advancing Human-Robot Collaboration for Industry 5.0: A Simulation-Based Hybrid Deep Learning and Reinforcement Learning Framework for Adaptive Task Allocation" by Claudio Urrea, submitted to Systems (MDPI, 2025). These artifacts include datasets, MATLAB scripts, tables, and figures to support the analysis of results presented in the paper (Section 4, Tables 2–3, Figures 2–6).

Overview

The artifacts support a simulation-based Hybrid Deep Learning and Reinforcement Learning (DL-RL) framework for adaptive task allocation in Human-Robot Collaboration (HRC). Implemented in MATLAB R2025a and RoboDK 5.9, the framework integrates a Convolutional Neural Network (CNN) for human state perception with a Double Deep Q-Network (DDQN) for dynamic task allocation. The simulation evaluates 1,000 HRC episodes, achieving improvements in throughput, workload, and safety compared to rule-based and SARSA baselines.

Repository Structure

HRCSimulationArtifacts/ ├── README.md # This file ├── Metadata.md # Detailed file descriptions and dependencies ├── PaperRoboDK.pdf # Full manuscript ├── Dataset/ │ ├── Results/ │ │ ├── HRCSimulationResults.csv # Performance metrics for 1,000 episodes │ │ ├── HRCSimulationLogEpisodes.txt # Simulation log │ │ ├── HRCSyntheticDataset.parquet # Synthetic dataset (~100 KB) │ ├── Tables/ │ │ ├── Table2.csv # Performance metrics for methods (Table 2) │ │ ├── Table3.csv # Metrics by human state (Table 3) │ ├── Figures/ │ │ ├── DebugFigure4.txt # Debug log for Figure 4 │ │ ├── DebugFigure6.txt # Debug log for Figure 6 ├── Scripts/ │ ├── hrcsimulation.m # Main simulation script │ ├── Table2.m # Generates Table 2 │ ├── Table3.m # Generates Table 3 │ ├── Figure2.m # Placeholder for Figure 2 (RoboDK snapshot) │ ├── Figure3.m # Generates Figure 3 (Q-table heatmap) │ ├── Figure4.m # Generates Figure 4 (conditional heatmaps) │ ├── Figure5.m # Generates Figure 5 (dynamic task allocation) │ ├── Figure6.m # Generates Figure 6 (proof-of-concept) │ ├── GenerateSyntheticDataset.m # Generates Parquet dataset ├── Figures/ │ ├── Figure1.tiff # System diagram illustrating the hybrid framework │ ├── Figure2.tiff # RoboDK snapshot (placeholder) │ ├── Figure3.tiff # Q-table heatmap │ ├── Figure4.tiff # Conditional heatmaps │ ├── Figure5.tiff # Dynamic task allocation │ ├── Figure6.tiff # Proof-of-concept visualization ├── SimulationAssets/ │ ├── PlaceholderMetadata.txt # Notes missing RoboDK templates, STEP/URDF models

Installation

1. MATLAB R2025a:

   • Install MATLAB R2025a with the following toolboxes:
     • Reinforcement Learning Toolbox
     • Deep Learning Toolbox
     • Robotics System Toolbox
     • Statistics and Machine Learning Toolbox
   • Recommended hardware: NVIDIA RTX 4090 GPU for CNN training.

2. RoboDK 5.9 (Optional):

   • Install RoboDK 5.9 for visualization (Section 2.1).
   • Set the API path to C:\RoboDK\API\MATLAB (modify as needed).
   • Note: Scripts run without RoboDK if connection fails (see robodkAvailable flag in hrc_simulation.m).

3. Repository Setup:

   • Clone this repository: ```bash git clone https://github.com/ClaudioUrrea/hrc-simulation.git

Alternatively, download files from FigShare ([DOI to be provided]). Place all files in a MATLAB-accessible working directory. Ensure HRCSimulationResults.csv is in the same directory as scripts.

Usage 1. Run the Simulation: - Execute hrcsimulation.m to simulate 1,000 HRC episodes. - Outputs: - HRCSimulation_Results.csv: Performance metrics. - Plots: Throughput, workload, and safety trends (saved as PNGs if exported). - Note: RoboDK visualization requires a valid connection.

1. Generate Tables:

   • Run Table2.m to produce Table2.csv (Section 4.1, Table 2).
   • Run Table3.m to produce Table3.csv (Section 4.1, Table 3).
   • Inputs: HRCSimulationResults.csv.

2. Generate Figures:

   • Run Figure_1.m for the System diagram illustrating the hybrid framework (Section 2, Figure 1).
   • Run Figure_2.m for a placeholder RoboDK snapshot (Section 3.1, Figure 2).
   • Run Figure_3.m for the Q-table heatmap (Section 4, Figure 3).
   • Run Figure_4.m for conditional heatmaps (Section 4, Figure 4).
   • Run Figure_5.m for dynamic task allocation (Section 4, Figure 5).
   • Run Figure_6.m for proof-of-concept visualization (Section 4, Figure 6).
   • Outputs: PNG files in Figures/ directory.
   • Note: Figures 4–6 require HRCSimulationResults.csv and/or qTable from hrc_simulation.m.

3. Generate Synthetic Dataset:

   • Run generatesyntheticdataset.m to create generatesyntheticdataset.parquet (Section 2.2).
   • Input: HRCSimulationResults.csv.

4. View Results:

   • Open HRCSimulationResults.csv, Table2.csv, Table3.csv in a spreadsheet or MATLAB.
   • View figures in Figures/ using an image viewer.
   • Check debug logs in Dataset/Figures/ for Figure 4 and Figure 6 details.

Metadata See Metadata.md for detailed file descriptions, column definitions (e.g., StateIndex = 1–9 for fatigue/skill combinations), and dependency information.

License All files are licensed under Creative Commons Attribution 4.0 International (CC-BY 4.0), as per the paper’s open-access statement.

Citation Urrea, C. (2025). Advancing Human-Robot Collaboration for Industry 5.0: A Simulation-Based Hybrid Deep Learning and Reinforcement Learning Framework for Adaptive Task Allocation. Systems, 13, [page range]. https://doi.org/10.3390/xxxxx.

Links • Dataset: FigShare ([DOI to be provided]) • Code: GitHub (https://github.com/ClaudioUrrea/hrc-simulation) • Preregistration: Open Science Framework ([URL to be provided])

Contact For questions or possible missing files, contact Claudio Urrea at claudio.urrea@usach.cl