---
license: other
pretty_name: CSRL simulator data
tags:
- robotics
- reinforcement-learning
- mujoco
---

# CSRL simulator data

Data a fresh checkout of the CSRL simulator needs (repo `26-sgrl-hhi`, branch `csrl-api`;
going forward `26-csrl-robohub`):

- `configs/` stable-state pools: the start/goal configurations of each scenario
- `models/` the trained TD7 experts (full checkpoints), one per scenario, as referenced by
  `files.<scenario>.agent` in the simulator's `config.yaml`
- `assets/` meshes the scenes load (Panda, Reachy Mini, Trossen)
- `mjcf/` MuJoCo exports of every scenario (what runs without rai)

Download into the simulator checkout:

```bash
hf download Shcherba/csrl-robohub-data --repo-type dataset --local-dir data
```

Used by "Estimate, don't imitate" (ICRA 2026; code: `HHI-Robot-Learning/26-DistillingSGRL`).
The robot meshes keep the licenses of their original sources.