test: add E2 MLP Core4 model card

e2_mlp_core4_current_state_20260910/README.md

@@ -0,0 +1,72 @@
+---
+license: apache-2.0
+tags:
+  - robotics
+  - imitation-learning
+  - dexflow
+  - residual-mlp
+  - dexterous-manipulation
+library_name: pytorch
+---
+
+# DexFlow MLP Core4 Stage-2 (E2)
+
+Human HandFlow + current object-state pretrain → robot 41D Residual MLP.
+
+Org: [Humantwin](https://huggingface.co/Humantwin)  
+Repo path on Hub: `e2_mlp_core4_current_state_20260910/`
+
+**Not ACT / DiT.** `DexFlowMLPPolicy` (`policy_variant=dexflow_mlp`).
+
+## Architecture
+
+```text
+RGB 480×640 → act41_resize_pad 224 + ImageNet
+  → ResNet18 → Z_int [8×256] → Residual MLP → action chunk [30, 41]
+```
+
+- `n_obs=1`, chunk 30 @ 30 Hz, no temporal ensemble, RTC off
+- Deploy: publish 30/30, infer interval ≥1.05 s
+- Normalization: per-task MEAN_STD `norm_stats.json` in the same folder
+- Transfer: backbone 100 + interaction 42; action head 0
+
+## Checkpoints (30k, G2 PASS)
+
+Same 10-ep robot set per task. Stage-1: `dexflow_mlp_stage1_core4_current_state_gate_20260910/stage1_step02000.ckpt`. Seed `20260910`.
+
+| Task dir | Robot task | Open-loop nRMSE |
+|---|---|---:|
+| `plastic/` | bottle_upright | 0.0458 |
+| `pill/` | med_bottle_into_box | 0.0599 |
+| `open/` | open_box_lid | 0.0508 |
+| `realsense/` | realsense_box_into_box | 0.0594 |
+
+Each folder:
+
+```text
+{task}/stage2.ckpt
+{task}/norm_stats.json
+```
+
+## Load
+
+```python
+from deploy.real_robot.ablation_e2_interaction_pretrain import load_task
+
+model, normalizer = load_task("e2_mlp_core4_current_state_20260910", "plastic", device="cuda")
+# image [B,1,3,480,640] float 0-1; state [B,41] MEAN_STD-normalized
+chunk_n = model.predict_action_chunk({"image": img, "state": state_n})
+action = normalizer.unnormalize_action(chunk_n)  # [B,30,41] raw
+```
+
+Or:
+
+```python
+from deploy.real_robot.load_p0_policy import load_dexflow_policy
+from dexflow.data.normalize import ActionStateNormalizer
+
+model = load_dexflow_policy("plastic/stage2.ckpt", device="cuda")
+normalizer = ActionStateNormalizer.load("plastic/norm_stats.json")
+```
+
+Code pack: `dexflow_mlp_infer_20260914.tar.zst` (model + inference, no weights).