TaCauchy: An Extensible FEM Framework for Vision-Based Tactile Simulation

TaCauchy is an extensible finite-element framework for vision-based tactile simulation, built on Isaac Sim / Isaac Lab and UIPC. It extends the TacEx codebase with direct Cauchy stress extraction, contact traction recovery, adaptive mesh refinement, and multi-sensor support (GelSight Mini, DIGIT, 9DTact).

Paper: TaCauchy: An Extensible FEM Framework for Vision-Based Tactile Simulation — accepted at IROS 2026. arXiv: 2606.20426.

Highlights

• Physics-first force fields: Cauchy stress from hyperelastic FEM, projected to contact surfaces for normal pressure and tangential traction
• Modular VBTS interface: plug-in optical backends (Taxim, FOTS, ManiSkill-style FEM rendering)
• Multi-sensor demos: unified UIPC backend for GelSight Mini, DIGIT, and 9DTact
• Isaac Lab integration: RL tasks, benchmarking, and interactive shape-touch demos

Quick Start

See REPRODUCTION.md for the full installation and reproduction guide.

git clone --recurse-submodules https://github.com/figsama/TaCauchy.git
cd TaCauchy
git lfs install
./scripts/setup_assets.sh          # pull and verify Git LFS assets
./scripts/setup_libuipc_patches.sh # apply TaCauchy's local libuipc changes
conda activate env_isaaclab        # after Isaac Sim + Isaac Lab setup
./tacex.sh -i
pip install -e source/tacex_uipc -v
python scripts/demos/shape_touch/simple_tactile_demo.py --sensor gelsight

Repository Layout

Path	Description
source/tacex/	VBTS sensor abstraction and optical simulators
source/tacex_uipc/	UIPC FEM integration, stress/contact extraction, mesh generation
source/tacex_assets/	Sensor and robot configuration plus Git LFS USD/calibration assets
source/tacex_tasks/	Isaac Lab RL environments
scripts/demos/shape_touch/	Main TaCauchy demos — force-field visualization and benchmarks

Main Demos

# Minimal demo: interpolated 20×20 force grid + tactile RGB
python scripts/demos/shape_touch/simple_tactile_demo.py --sensor gelsight

# Full modular visualization
python scripts/demos/shape_touch/contact_force_visualization_demo_modular.py

# Performance benchmark (headless)
python scripts/demos/shape_touch/benchmark_tactile_performance.py --sensor gelsight --headless

Sensor options: --sensor gelsight | digit | 9dtact

Documentation

• REPRODUCTION.md — installation & reproduction
• ASSETS.md — large binary assets policy
• docs/source/installation/Local-Installation.md — detailed TacEx/UIPC install

Citation

If you use TaCauchy, please cite our paper (IROS 2026) and acknowledge TacEx:

@inproceedings{zhao2026tacauchy,
  title     = {TaCauchy: An Extensible FEM Framework for Vision-Based Tactile Simulation},
  author    = {Hengfei Zhao and Yifan Xie and Junhao Gong and Yue Sun and Kai Zhu and Weihua He and Shoujie Li and Haohuan Fu and Wenbo Ding},
  booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
  year      = {2026},
  eprint    = {2606.20426},
  archivePrefix = {arXiv},
  primaryClass  = {cs.RO},
  url       = {https://arxiv.org/abs/2606.20426}
}

@article{nguyen2024tacex,
  title   = {TacEx: GelSight Tactile Simulation in Isaac Sim -- Combining Soft-Body and Visuotactile Simulators},
  author  = {Duc Huy Nguyen and Tim Schneider and Guillaume Duret and Alap Kshirsagar and Boris Belousov and Jan Peters},
  year    = {2024},
  eprint  = {2411.04776},
  archivePrefix = {arXiv},
  primaryClass  = {cs.RO}
}

Acknowledgements

TaCauchy builds upon TacEx, Isaac Lab, Taxim, FOTS, and UIPC / libuipc.

License

MIT License — see upstream TacEx and libuipc for component-specific licenses.