BIue-Jay commited on
Commit
28ae760
Β·
verified Β·
1 Parent(s): d0c44bc

Update model card (task / checkpoint table)

Browse files
Files changed (1) hide show
  1. README.md +47 -27
README.md CHANGED
@@ -6,30 +6,50 @@ tags:
6
  ---
7
  # GR00T N1.7 β€” Humanoid Benchmark
8
 
9
- GR00T N1.7 checkpoints fine-tuned on Humanoid Benchmark tasks (2026-11-11 training round: global batch 696 = 87 per GPU
10
- x 8 GPUs, one node per run, 20 epochs).
11
-
12
- | Folder | Task | Robot |
13
- |---|---|---|
14
- | `b6-pedal-bin/` | Pick up a chip barrel, step on the pedal bin, drop it in | Unitree G1 |
15
- | `a3-fridge-open/` | Open the fridge door | Unitree G1 |
16
- | `a7-cabinet-close/` | Close the cabinet door | Unitree G1 |
17
- | `a18-pour-water/` | Pour water from a pitcher into a glass | H2 (Sharpa hands) |
18
- | `a10-trolley-push/` | Push a trolley | Unitree G1 |
19
- | `a20-basket-transfer/` | Transfer a basket | Unitree G1 |
20
- | `a17-unscrew-bottle/` | Unscrew the cap of a bottle | H2 (Sharpa hands) |
21
- | `b7-black-clothes-hamper/` | Take the black garment off the hanger and drop it into the hamper | Unitree G1 |
22
-
23
- Each task folder has `epoch-010/` and `epoch-020/` (100 demonstrations per task, 20 epochs, saved at epoch 10 and 20).
24
- Every epoch folder is self-contained for deployment and evaluation: weights, model / processor configuration,
25
- normalization statistics and `protocol.json` (input/output dimensions, cameras, step counts, file list).
26
- Training records are included for reproducibility (`training_plan.json`, `train.log`, and for GR00T `trainer_state.json`,
27
- `training_args.bin`, `wandb_config.json`); optimizer / scheduler / RNG / DeepSpeed resume state is not. `code/` holds the adapter code used for training and inference.
28
-
29
- Model inputs are identical across all benchmark models: G1 state 44-D (29 body joint angles + 6 + 6 measured hand
30
- values + 3 projected gravity), H2 state 78-D (75 joint angles + 3 projected gravity), no joint velocities.
31
- Training used the Humanoid Benchmark online data augmentation spec v2 (photometric, geometric, sensor and lighting
32
- augmentation plus proprioception noise/delay/drop), with the model's own image augmentation turned off.
33
- The VLM backbone `nvidia/Cosmos-Reason2-2B` is loaded from its own (gated) repository.
34
-
35
- The previous training round of this repository is kept in the git history at revision `2e8f375584631cae0454911e58b5b7850ea9f678` (older batch setting).
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6
  ---
7
  # GR00T N1.7 β€” Humanoid Benchmark
8
 
9
+ GR00T N1.7 (Gr00tN1d7) checkpoints fine-tuned on Humanoid Benchmark tasks (action head fine-tuned, VLM backbone frozen). The VLM backbone `nvidia/Cosmos-Reason2-2B` is loaded from its own (gated) repository.
10
+
11
+ ## Checkpoints
12
+
13
+ 100 demonstrations per task. Every row is one training run; the epoch folders are its published checkpoints.
14
+
15
+ | Folder | Task | Robot | Cameras (model input) | Global batch | Proprio augmentation | Checkpoints |
16
+ |---|---|---|---|---|---|---|
17
+ | `b6-pedal-bin/` | B6 β€” pick up a chip barrel, step on the pedal bin, drop it in | Unitree G1 (Inspire hands) | ego_view, chest | 696 | aligned (GR00T A20 recipe) | `epoch-010`, `epoch-020` |
18
+ | `a3-fridge-open/` | A3 β€” open the fridge door | Unitree G1 (Inspire hands) | ego_view, chest | 696 | aligned (GR00T A20 recipe) | `epoch-010`, `epoch-020` |
19
+ | `a3-fridge-open/b256/` | A3 β€” open the fridge door (re-train, current protocol) | Unitree G1 (Inspire hands) | ego_view, chest | 256 | aligned (GR00T A20 recipe) | `epoch-010`, `epoch-020` |
20
+ | `a7-cabinet-close/` | A7 β€” close the cabinet door | Unitree G1 (Inspire hands) | ego_view, chest | 696 | aligned (GR00T A20 recipe) | `epoch-010`, `epoch-020` |
21
+ | `a7-cabinet-close/b256/` | A7 β€” close the cabinet door (re-train, current protocol) | Unitree G1 (Inspire hands) | ego_view, chest | 256 | aligned (GR00T A20 recipe) | `epoch-010`, `epoch-020` |
22
+ | `a6-door-pull/` | A6 β€” pull a door open | Unitree G1 (Inspire hands) | ego_view, chest | 256 | aligned (GR00T A20 recipe) | `epoch-010`, `epoch-020` |
23
+ | `a10-trolley-push/` | A10 β€” push a trolley | Unitree G1 (Inspire hands) | ego_view, chest | 696 | aligned (GR00T A20 recipe) | `epoch-010`, `epoch-020` |
24
+ | `a20-basket-transfer/` | A20 β€” transfer a basket | Unitree G1 (Inspire hands) | ego_view, chest | 696 | aligned (GR00T A20 recipe) | `epoch-010`, `epoch-020` |
25
+ | `b7-black-clothes-hamper/` | B7 β€” take the black garment off the hanger and drop it into the hamper | Unitree G1 (Inspire hands) | ego_view, chest | 696 | aligned (GR00T A20 recipe) | `epoch-010`, `epoch-020` |
26
+ | `c6-bathroom-door-fetch/` | C6 β€” open the bathroom door, take the tea bottle from the vanity, put it on the dining table | Unitree G1 (Inspire hands) | ego_view, chest | 256 | aligned (GR00T A20 recipe) | `epoch-010`, `epoch-020` |
27
+ | `a18-pour-water/` | A18 β€” pour water from a pitcher into a glass | H2 (Sharpa hands) | head, chest | 696 | aligned (GR00T A20 recipe) | `epoch-010`, `epoch-020` |
28
+ | `a17-unscrew-bottle/` | A17 β€” unscrew the cap of a bottle | H2 (Sharpa hands) | head, chest | 696 | aligned (GR00T A20 recipe) | `epoch-010`, `epoch-020` |
29
+
30
+ A `b256/` or `aligned/` sub-folder is a re-train of that task under the current benchmark protocol (2026-11-12: global
31
+ batch GR00T / pi0.5 / Delta Flow / ACT 256, Psi-Zero 512, DiT4DiT 64; proprioception augmentation aligned with the
32
+ GR00T A20 recipe) and is the one to use for that task; the checkpoints directly under the task folder are the earlier
33
+ training round, kept for reference.
34
+
35
+ ## What an epoch folder contains
36
+
37
+ Every `epoch-XXX/` folder is self-contained for evaluation and deployment: weights, model configuration, normalization
38
+ statistics and `protocol.json` (task, data, input/output dimensions, cameras, batch, steps, augmentation, file list).
39
+ Simple training records are included for reproducibility (training plan, training log, configs, metrics); optimizer /
40
+ scheduler / RNG / DeepSpeed resume state is not. `code/` holds the adapter code used for training and inference
41
+ (`code/VERSIONS.json`: upstream commits and package versions).
42
+
43
+ ## Inputs and augmentation
44
+
45
+ Model inputs are identical across all benchmark models: G1 state 44-D (29 body joint angles + 6 + 6 measured hand values
46
+ + 3 projected gravity), H2 state 78-D (75 joint angles + 3 projected gravity), no joint velocities. Cameras: G1 ego view +
47
+ chest; H2 head + chest (A14: head + chest + wrist). The third-person camera recorded in the datasets is for data review
48
+ only and is never a model input. Each camera is a separate image input, except for Delta Flow and DiT4DiT, which
49
+ place the views side by side in one frame.
50
+ Training uses the Humanoid Benchmark online data augmentation spec v2 (15 photometric / geometric / sensor / lighting
51
+ effects, one of them boosted to 1.5x per sample) with the model's own image augmentation turned off, plus joint-angle
52
+ bias and noise on the proprioception (column "Proprio augmentation"; the earlier recipe also delayed the
53
+ proprioception and dropped action-history rows).
54
+
55
+ Earlier repository states are kept in the git history (e.g. revision `2e8f375584631cae0454911e58b5b7850ea9f678`, an older batch setting).