---
license: apache-2.0
task_categories:
- robotics
tags:
- LeRobot
- SIMPLE
- humanoid
- unitree-g1
- memory
configs:
- config_name: default
  data_files: data/*/*.parquet
---

# SimFactoryFlashPadSingle-v0

Memory-dependent humanoid manipulation demonstrations for the Unitree G1, in the same LeRobot v2.1 layout as SIMPLE's `G1WholebodyTabletopGraspMP-v0`.

**Task.** Watch the lamp on the table; it flashes one, two or three times. Then pick up the apple and place it on pad 1, 2 or 3, counting the pads from the left, matching the number of flashes.

The flashes are brief (0.5 s on, 1.0 s off) and all happen before the robot moves. The robot's motion until the pad choice is the same for every count, so the policy must remember how many flashes it saw.

## Contents

| Item | Value |
|---|---|
| Episodes | 119 (40 × 1 flash, 40 × 2 flashes, 39 × 3 flashes) |
| Frames | 310,990 at 50 fps (episodes 42–105 s, median 48 s) |
| Camera | `observation.images.egocentric` (G1 `head_stereo_left`), 640x360, h264 |
| `states` / `action` | 32-d / 36-d, same semantics as SIMPLE (see `meta/modality.json`) |
| Format | LeRobot codebase v2.1 |

## Generation

- Physics: MuJoCo, scripted expert (SIMPLE G1 controller): validated front-on apple pick, pick-style approach to the remembered pad, table placement.
- Rendering: NVIDIA Isaac Sim replay of the recorded trajectories.
- Domain randomization per episode, SIMPLE style: lighting, table, ground and object materials, and 1 to 3 render-only GraspNet distractors.
- Curation: only complete episodes with the apple within 8 cm of the correct pad and nearer to it than any other pad (all kept episodes are within 4 mm). Every flash is verified in the recorded egocentric frames: the lit lamp is present in every frame of every flash window.

## Files

- `data/`, `meta/`, `videos/`: the LeRobot v2.1 dataset. Per-episode `environment_config.memory.flash_count` holds the answer.
- `simple/SimFactoryFlashPadSingle-v0.zip`: the same files as one archive, in the layout of `USC-PSI-Lab/psi-data` (`simple/<task>.zip`). Ψ₀ training tools download this archive at a pinned revision.
- `meta/stats_psi0.json`: the normalization statistics Ψ₀ uses, identical to `meta/stats.json`.

## Loading

```python
from lerobot.datasets.lerobot_dataset import LeRobotDataset
ds = LeRobotDataset("randing2000/SimFactoryFlashPadSingle-v0")
```