Add dataset conversion configuration
Browse files
unitreerobotics_G1_Dex3_BlockStacking_Dataset.yaml
ADDED
|
@@ -0,0 +1,47 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
hf_source: "unitreerobotics/G1_Dex3_BlockStacking_Dataset"
|
| 2 |
+
subset: null
|
| 3 |
+
split: "train"
|
| 4 |
+
revision: "main"
|
| 5 |
+
target_name: "unitreerobotics_G1_Dex3_BlockStacking_Dataset"
|
| 6 |
+
|
| 7 |
+
conversion:
|
| 8 |
+
mode: "script"
|
| 9 |
+
converter: "unitreerobotics_g1_dex3_blockstacking"
|
| 10 |
+
input_format: "parquet"
|
| 11 |
+
time_column: "timestamp"
|
| 12 |
+
time_precision: "ms"
|
| 13 |
+
tag_columns: ["episode_index", "task_index"]
|
| 14 |
+
rename: {}
|
| 15 |
+
drop: []
|
| 16 |
+
script_params:
|
| 17 |
+
source_dir: "D:\\code\\py\\unitreerobotics_G1_Dex3_BlockStacking_Dataset"
|
| 18 |
+
table_name: "unitreerobotics_g1_dex3_blockstacking"
|
| 19 |
+
fps: 30
|
| 20 |
+
time_source: "timestamp"
|
| 21 |
+
granularity: "merged"
|
| 22 |
+
tsfile_writer: "python"
|
| 23 |
+
stream_staging: false
|
| 24 |
+
encoding:
|
| 25 |
+
float: "GORILLA"
|
| 26 |
+
double: "GORILLA"
|
| 27 |
+
int32: "TS_2DIFF"
|
| 28 |
+
int64: "TS_2DIFF"
|
| 29 |
+
time: "TS_2DIFF"
|
| 30 |
+
boolean: "RLE"
|
| 31 |
+
compression:
|
| 32 |
+
default: "LZ4"
|
| 33 |
+
time: "LZ4"
|
| 34 |
+
|
| 35 |
+
upload:
|
| 36 |
+
mode: "single"
|
| 37 |
+
repo_id: null
|
| 38 |
+
|
| 39 |
+
attribution:
|
| 40 |
+
original_url: "https://huggingface.co/datasets/unitreerobotics/G1_Dex3_BlockStacking_Dataset"
|
| 41 |
+
license: "apache-2.0"
|
| 42 |
+
authors: "Unitree Robotics (dataset repository owner); source tree contributors Henry-Ellis and lv-1"
|
| 43 |
+
notes: |
|
| 44 |
+
The source is a 30 Hz LeRobot dataset with 301 episodes and 281,196 rows.
|
| 45 |
+
Time is round(timestamp * 1000) in milliseconds and restarts at zero per episode.
|
| 46 |
+
observation.state and action are flattened from 28-element float32 vectors.
|
| 47 |
+
Source videos are retained only at the original Hugging Face videos/ path.
|