File size: 2,813 Bytes
c1a57d2
 
 
 
 
33712e7
c1a57d2
 
 
90ebe8a
c1a57d2
 
 
 
 
 
 
 
33712e7
 
 
c1a57d2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
---
license: apache-2.0
task_categories:
- robotics
tags:
- LeRobot
- tsfile
- timeseries
- robotics
- format:tsfile
pretty_name: NVIDIA WORKSPACE
size_categories:
- 10K<n<100K
configs:
- config_name: default
  data_files:
  - split: train
    path: data/nvidia_workspace_train.tsfile
modality:
- tabular
- timeseries
---

# NVIDIA WORKSPACE (TsFile)

Apache TsFile version of
[`tremmelnicholas/NVIDIA_WORKSPACE`](https://huggingface.co/datasets/tremmelnicholas/NVIDIA_WORKSPACE).

## Overview

This LeRobot v3.0 dataset records an SO-101 follower robot performing the task
"Pick up the vial and place it in the rack."

- **Scale:** 27 episodes, 19,913 frames, one task, and one train split.
- **Sampling rate:** 30 Hz.
- **Modalities:** Time-series. The source dataset also provides video.
- **Source cameras:** ego and external Intel RealSense D455 streams at
  640 x 480, encoded as AV1 video at 30 fps.
- **Artifact:** one TsFile, `data/nvidia_workspace_train.tsfile`.

## TsFile schema

- **Time** (INT64, milliseconds): `round(timestamp * 1000)`, restarting for
  each episode.
- **episode_index**, **task_index** (TAG): episode and task dimensions. Use the
  TAG columns to select one trajectory, for example `episode_index=0`.
- **frame_index**, **sample_index** (FIELD, INT64): the frame position within
  an episode and the renamed global source `index`.
- **observation_state_0..observation_state_5** (FIELD, FLOAT): shoulder pan,
  shoulder lift, elbow flex, wrist flex, wrist roll, and gripper positions.
- **action_0..action_5** (FIELD, FLOAT): commands for the same six robot axes.

The source `observation.state[6]` and `action[6]` vectors are flattened without
discarding elements. The source `timestamp` field is represented by `Time` and
is not duplicated as a FIELD. Source `index` is renamed to `sample_index`.

The two video features, `observation.images.ego` and
`observation.images.external_D455`, are not uploaded here. Their MP4 files
remain available in the
[original videos directory](https://huggingface.co/datasets/tremmelnicholas/NVIDIA_WORKSPACE/tree/main/videos).
The mirrored `meta/info.json` records the flattened columns, time mapping,
omitted video features, and row count.

## Usage

```python
from tsfile import TsFileReader

reader = TsFileReader("data/nvidia_workspace_train.tsfile")
schemas = reader.get_all_table_schemas()
print(schemas)
reader.close()
```

All episodes share the `nvidia_workspace_train` table. Query by
`episode_index` and `task_index` to isolate a trajectory.

## Source and license

- Original dataset: https://huggingface.co/datasets/tremmelnicholas/NVIDIA_WORKSPACE
- Author / publisher: tremmelnicholas
- Framework: LeRobot v3.0
- Robot type: `so101_follower`
- Paper or homepage: not declared by the source dataset
- License: Apache License 2.0