Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,26 @@
|
|
| 1 |
---
|
| 2 |
license: mit
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: mit
|
| 3 |
---
|
| 4 |
+
|
| 5 |
+
# DISC: Decoupling Instruction from State-Conditioned Control via Policy Generation
|
| 6 |
+
|
| 7 |
+
Official implementation of the RSS 2026 paper:
|
| 8 |
+
|
| 9 |
+
**DISC: Decoupling Instruction from State-Conditioned Control via Policy Generation**
|
| 10 |
+
|
| 11 |
+
> Hanxiang Ren, Pei Zhou, Xunzhe Zhou, and Yanchao Yang
|
| 12 |
+
> Robotics: Science and Systems (RSS), 2026
|
| 13 |
+
|
| 14 |
+
## Overview
|
| 15 |
+
|
| 16 |
+
DISC is a policy-generation framework that decouples high-level language instructions from state-conditioned low-level control. This repository provides training, evaluation, and test-time training code for reproducing the results reported in the paper.
|
| 17 |
+
|
| 18 |
+
## Content
|
| 19 |
+
|
| 20 |
+
The repo currently contains two checkpoints to reproduce the results reported in the paper.
|
| 21 |
+
|
| 22 |
+
- `DISC-libero90.ckpt` contains the model trained on LIBERO-90 dataset (Table 1)
|
| 23 |
+
|
| 24 |
+
- `DISC-libero_cotrain_spatial_object_goal.ckpt` contains the model cotrained on LIBERO-sptial, LIBERO-object, LIBERO-goal. (Table 2)
|
| 25 |
+
|
| 26 |
+
More checkpoints will be uploaded later.
|