---
license: mit
task_categories:
  - reinforcement-learning
tags:
  - robomimic
  - offline-rl
  - ope
---

# Robomimic mhf dense datasets

Mixed-human/filtered (`mhf`) dense-reward Robomimic datasets used by infOPE / VGPS experiments.

## Datasets

| Path | Successful / total trajectories |
|---|---|
| `lift/mhf_densev1` | 202 / 300 |
| `lift/mhf_densev1_ntrajs=1020` | 775 / 1020 |
| `can/mhf_densev1` | 210 / 300 |
| `square/mhf_densev1` | 48 / 300 |
| `square/mhf2_densev1` | 207 / 300 |

## Layout

```
lift/
  mhf_densev1/dataset.hdf5
  mhf_densev1_ntrajs=1020/dataset.hdf5
can/
  mhf_densev1/dataset.hdf5
square/
  mhf_densev1/dataset.hdf5
  mhf2_densev1/dataset.hdf5
```

## Download

```bash
# full repo
hf download infope/robomimic --repo-type dataset --local-dir ./robomimic

# one variant
hf download infope/robomimic lift/mhf_densev1/dataset.hdf5 --repo-type dataset
```