Upload FunctionalHammerStrike_v1/shadow_right-best-44/README.md with huggingface_hub
Browse files
FunctionalHammerStrike_v1/shadow_right-best-44/README.md
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# FunctionalHammerStrike_v1/shadow_right-best-44
|
| 2 |
+
|
| 3 |
+
State-input Diffusion Policy (no images) on **Dexverse-FunctionalHammerStrike-v1**, hand **shadow_right**, author v1 demos, epoch **44** / max 300
|
| 4 |
+
(early stopping on validation loss, patience 40); role **best** = lowest validation loss.
|
| 5 |
+
|
| 6 |
+
- state: previous action + joints + object state + goal + contact (83 dims), obs / action horizon 2 / 16
|
| 7 |
+
- U-Net diffusers (256, 512, 1024), batch 256, lr 0.0001
|
| 8 |
+
- inference: DDPM 20 steps, 16-step chunks (`scripts/diffusion/eval_online.py`)
|
| 9 |
+
- training run: `20261015-111200_a4_dp-v1-functionalhammerstrike` (hscho_share experiments/a4/dexverse-dp-no-images)
|