Time Consistent Surface Mapping for Deformable Object Shape Control

This repository contains the MATLAB code and a Python implementation for replicating the surface mapping results presented in our paper:

[1] I. Cuiral-Zueco and G. López-Nicolás, "Time Consistent Surface Mapping for Deformable Object Shape Control," IEEE Transactions on Automation Science and Engineering, doi: 10.1109/TASE.2025.3529180.

To access the paper and the user-friendly paper web page https://nachocz.github.io/time-consistent-surface-mapping/

---

Disclaimer

The .mat data files in this repository are stored using Git LFS. To properly access the data, clone the repository using:

git clone https://github.com/nachocz/time-consistent-surface-mapping

Do not download the repository as a ZIP, as this will not generate proper .mat files.

The cloning process might take a few extra seconds, but after that, everything should work properly.

---

Overview

This code demonstrates the implementation of time-consistent surface mapping for deformable objects, a technique aimed at achieving accurate and efficient shape control. The repository includes scripts to reproduce experimental results, including:

• Surface maps: Visualization of the deformation mapping process.
• Shape error analysis: Quantitative comparison of shape errors.
• Time cost analysis: Evaluates performance, which may vary depending on your system's specifications.
• Mesh size evaluation: Examination of the generated meshes.

Example plots generated by the script timeconsistentsurfacemapping.m include surface maps, shape error graphs, time cost comparisons, and mesh size plots:

---

Repository contents

The repository includes:

• python/: Clean Python implementation with interactive 3D visualisation (see python/README.md).

• data/: Contains experiment videos and mesh data for analysis.

  • .mp4 files: Videos of the shape control experiments.
  • .mat files: Mesh data generated from colour-segmented (CieLAB space) RGB-D frames in the experiments.

• utils/: Utilities for mesh analysis, obtained from https://github.com/llorz/SGA19_zoomOut.

• timeconsistentsurfacemapping.m: Main script to replicate the surface mapping results presented in the paper.

• objectsurfacemeshing.m: Script illustrating the conversion of raw point cloud data into the mesh structures stored in the data folder.

---

Dependencies

To run the scripts, you need MATLAB installed with the following toolboxes:

• Statistics and Machine Learning Toolbox
• Computer Vision Toolbox
• Lidar Toolbox
• Optimization Toolbox

---

Features and capabilities

The code provides the following functionalities:

1. Time-Consistent Surface Mapping: Implements a method to achieve consistent deformation mapping over time for deformable objects.
2. Performance Comparison: Benchmarks the proposed method against a classical ZoomOut approach.
3. Error and Time Analysis: Evaluates shape errors and computational time for each experiment.
4. Visualization: Generates plots for surface mapping, shape errors, time cost, and mesh sizes.

---

Experiment data

The data folder contains experiment data for five different deformable objects:

1. Mexican hat
2. T-shaped noodle
3. Pillow
4. Foam rectangle
5. Foam free-shape (custom foam cutout)

Each .mat file in the data folder corresponds to the RGB-D mesh data for one of these objects, enabling reproduction of the surface mapping results.

---

Python implementation

A standalone Python implementation is available in the python/ folder with:

• NumPy/SciPy-based functional map pipeline (cotangent Laplacian, ZoomOut, time-consistent refinement)
• Interactive 3D visualisation using Polyscope with live frame-by-frame playback
• Side-by-side comparison of ZoomOut vs. time-consistent mapping

cd python
pip install -r requirements.txt
python demo.py --experiments 1       # run with 3D viewer
python demo.py --no-gui              # headless mode

See python/README.md for full details.

---

Experimental setup

The experiments were conducted using a robotic setup designed for the shape control of deformable objects. The mesh data provided in the data folder was obtained using an RGB-D RealSense camera. The process of converting raw point cloud data into mesh structures is illustrated in the objectsurfacemeshing.m script.

Below is an image of the robotic setup used in the experiments:

---

Citation

The related paper can be cited as:

[1] I. Cuiral-Zueco and G. López-Nicolás, "Time Consistent Surface Mapping for Deformable Object Shape Control," IEEE Transactions on Automation Science and Engineering, doi: 10.1109/TASE.2025.3529180.

• Ignacio Cuiral-Zueco
• Gonzalo López-Nicolás

Acknowledgements

This work was supported through Project REMAIN S1/1.1/E0111 (Interreg Sudoe Programme, ERDF), Project PID2021-124137OB-I00, and Project TED2021-130224B-I00 funded in part by MCIN/AEI/10.13039/501100011033, in part by the ERDF A way of making Europe, and in part by the European Union NextGenerationEU/PRTR.