Datasets:
Update README.md
Browse files
README.md
CHANGED
|
@@ -9,14 +9,22 @@ configs:
|
|
| 9 |
data_files: data/*/*.parquet
|
| 10 |
---
|
| 11 |
|
| 12 |
-
|
|
|
|
| 13 |
|
| 14 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
|
|
|
|
| 16 |
|
|
|
|
|
|
|
|
|
|
| 17 |
|
| 18 |
-
- **Homepage:** [More Information Needed]
|
| 19 |
-
- **Paper:** [More Information Needed]
|
| 20 |
- **License:** apache-2.0
|
| 21 |
|
| 22 |
## Dataset Structure
|
|
|
|
| 9 |
data_files: data/*/*.parquet
|
| 10 |
---
|
| 11 |
|
| 12 |
+
The goal of this dataset is to record an SO101 arm while it picks up a matchbox, places
|
| 13 |
+
it on a black square and then goes into a resting position (fully closed).
|
| 14 |
|
| 15 |
+
The recording was done in a way to also support removing the matchbox from the black
|
| 16 |
+
square and putting it in front of the robot. There's also one sample without a matchbox
|
| 17 |
+
with no action required.
|
| 18 |
+
|
| 19 |
+
Light and camera angle varies quite a bit to test the generalization capability of the
|
| 20 |
+
policy.
|
| 21 |
|
| 22 |
+
We found that ACT learns a lot better than SmolVLA.
|
| 23 |
|
| 24 |
+
This dataset was created using [LeRobot](https://github.com/huggingface/lerobot).
|
| 25 |
+
|
| 26 |
+
## Dataset Description
|
| 27 |
|
|
|
|
|
|
|
| 28 |
- **License:** apache-2.0
|
| 29 |
|
| 30 |
## Dataset Structure
|