File size: 2,390 Bytes
23a5358
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
---
license: cc-by-4.0
task_categories:
- robotics
tags:
- LeRobot
- so101
- teleoperation
- quantization
- jetson
pretty_name: QuantGate SO-101 cola can to coaster (120 demos)
size_categories:
- 10K<n<100K
configs:
- config_name: default
  data_files: data/*/*.parquet
---

# QuantGate SO-101 demonstrations: cola can to coaster

These are 120 teleoperated demonstrations of one task on an SO-101 arm, recorded with LeRobot 0.6.1 (dataset format
v3.0). They are the training and audit data of **QuantGate**. QuantGate gates INT8, FP16 and FP32 TensorRT builds of
an ACT policy before they reach a Jetson Orin Nano Super driving the same arm.

- **Task**: "Pick up the cola can and place it upright on the coaster". The can is an unopened 268 g cola can. It
  starts at a marked spot and goes onto a fixed coaster.
- **Episodes**: 120. **Frames**: 65,424 at 30 fps. Episode length: 419–893 frames (median 523).
- **Robot**: SO-101 follower and leader arms (Feetech STS3215). State and action are six joint positions:
  `shoulder_pan, shoulder_lift, elbow_flex, wrist_flex, wrist_roll, gripper`.
- **Cameras** (AV1 video):
  - `observation.images.top`: Logitech C922 overhead, 640×480, with locked exposure and white balance.
  - `observation.images.wrist`: OpenMV AE3 wrist camera, 320×200.
- **All episodes are successful demonstrations.**

## Splits

`splits/lock1_split_manifest.yaml` assigns the episodes to **train 80 / val 20 / audit 20**. The split is seeded
(`PREREG-SEED-2026-09-14`) and was frozen before training (LOCK-1).
- The ACT policy was trained on `train` only.
- `audit` episodes were never used for training or calibration. They are the held-out states of the offline audit.
- `splits/lock1_dataset_files.json` holds the SHA-256 of every data file as frozen at LOCK-1, and this upload matches it.
- `checksums.sha256` covers every file in this repository.

## Use

```python
from lerobot.datasets.lerobot_dataset import LeRobotDataset
ds = LeRobotDataset("chans/quantgate-so101-coaster")
```

## People in the data

The operator's hand and arm can appear in the overhead view while they reset the scene. No faces are recorded, and
there is no audio.

## Related

- Companion dataset with varied start positions: `chans/quantgate-so101-coaster-positions`.
- Code, pre-registrations and results: the QuantGate repository (public from 2026-10-26).

License: CC BY 4.0.