Tactile Genesis

!! Please note, this is not the final version of the paper. We plan to update ALL experiments with latest version of Genesis World. !!

tactile_genesis_overview_compressed.mp4

This repository reflects the self-contained code snapshot to reproduce the paper, Tactile Genesis. Please note, if you wish to use tactile sensors for your own project, use the latest Genesis World physics engine which includes our tactile sensors, and optionally, the official Eden (not yet available, but will be released before end of 2026) if you require a managed learning framework.

Structure

Directory	Package	Role
dexterous-hands/	tactile_genesis	Tactile dexterous-hand RL tasks, robots, sensors, training/eval entrypoints.
Eden/	eden	RL environment framework built on Genesis + rsl_rl. Declarative Pydantic configs, manager/term registries.
Genesis/	genesis-world	The physics-simulation backend (rigid/MPM/SPH/FEM/PBD).
rsl_rl/	rsl-rl-lib	PPO + student/teacher distillation training loop.

Dependency topology: dexterous-hands → Eden → Genesis, with rsl_rl as the training loop.

How the packages are wired

dexterous-hands/pyproject.toml resolves the three dependencies from the sibling checkouts via [tool.uv.sources] (editable path installs). These sources override the package origin across the whole resolution graph, so Eden's own transitive genesis-world / rsl-rl-lib requirements also point at the local checkouts:

[tool.uv.sources]
eden          = { path = "../Eden",    editable = true }
genesis-world = { path = "../Genesis", editable = true }
rsl-rl-lib    = { path = "../rsl_rl",  editable = true }

Quickstart

Everything is run from dexterous-hands/:

cd dexterous-hands
uv sync  # installs eden, genesis-world, rsl-rl-lib from local
uv run python main.py --task=in_hand_repose --robot=xhand1 --cpu

See dexterous-hands/README.md for the full task/robot list, run modes, and tooling scripts.

Citation

If you use or reference our simulated tactile sensors in any way, please cite:

@article{chung2026tactilegenesis,
  title   = {Tactile Genesis: Exploring Tactile Sensors at Scale for Learning Dexterous Tasks},
  author  = {Chung, Trinity and Yamazaki, Kashu and Patel, Dhruv and Duburcq, Alexis and Qiao, Yiling and Fragkiadaki, Katerina and Nayebi, Aran},
  journal = {arXiv preprint arXiv:2606.22332},
  year    = {2026},
  url     = {https://arxiv.org/abs/2606.22332}
}