README.md

---
license: apache-2.0
library_name: jax
tags:
- robotics
- vision-language-action
- vla
- flow-matching
- manipulation
- garment-folding
- lehome-challenge
pipeline_tag: robotics
datasets:
- lehome/dataset_challenge_merged
---

# LeHome Challenge 2026 — Simulation Policy

This is my policy for the **online (simulation) round** of the [LeHome Challenge 2026](https://github.com/lehome-official/lehome-challenge) (ICRA 2026), a competition on bimanual garment folding with a LeRobot SO-ARM101 setup. It placed **1st of 62 teams**.

It is a π0.5-based vision-language-action policy trained with an asynchronous RL + DAgger loop entirely in NVIDIA Isaac Sim.

For how it works and how to run it, please refer to the code repository and the tech report.

- 💻 **Code:** https://github.com/IliaLarchenko/lehome_solution
- 📄 **Tech report:** https://arxiv.org/abs/2606.27163
- ✍️ **Blog post:** https://ilialarchenko.com/projects/lehome2026
- 🤖 **Real-world final policy:** https://huggingface.co/IliaLarchenko/lehome_real