Link the original DATAFARM paper and project
Browse files
README.md
CHANGED
|
@@ -17,6 +17,12 @@ Pretrained DATAFARM motion encoder weights converted losslessly to a pickle-free
|
|
| 17 |
NumPy archive for inference in JAX. The implementation and PyRoki integration are
|
| 18 |
in [DataFarm-PyRoki](https://github.com/abhaybd/DataFarm-PyRoki).
|
| 19 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 20 |
The frozen encoder maps a joint trajectory to a 16-dimensional latent mean. A
|
| 21 |
fixed DROID reference Gaussian supplies the squared Mahalanobis motion-style cost
|
| 22 |
used during trajectory optimization and subsequent stroke retiming.
|
|
|
|
| 17 |
NumPy archive for inference in JAX. The implementation and PyRoki integration are
|
| 18 |
in [DataFarm-PyRoki](https://github.com/abhaybd/DataFarm-PyRoki).
|
| 19 |
|
| 20 |
+
Original method: [DATAFARM: Distribution-Aligned Task and Motion Planning for
|
| 21 |
+
Fine-Tuning Vision-Language-Action Models](https://arxiv.org/abs/2609.12316)
|
| 22 |
+
by Samrat Sahoo, Yixuan Huang and Tom Silver (2026).
|
| 23 |
+
See the [original DATAFARM repository](https://github.com/SamratSahoo/DATAFARM)
|
| 24 |
+
and [project website](https://prpl-group.com/datafarm/).
|
| 25 |
+
|
| 26 |
The frozen encoder maps a joint trajectory to a 16-dimensional latent mean. A
|
| 27 |
fixed DROID reference Gaussian supplies the squared Mahalanobis motion-style cost
|
| 28 |
used during trajectory optimization and subsequent stroke retiming.
|