GR00T N1.7 — Humanoid Benchmark
GR00T N1.7 checkpoints fine-tuned on Humanoid Benchmark tasks (2026-11-10 training round, per-GPU batch 320, 32 GPUs).
| Folder | Task | Robot |
|---|---|---|
b6-pedal-bin/ |
Pick up a chip barrel, step on the pedal bin, drop it in | Unitree G1 |
a3-fridge-open/ |
Open the fridge door | Unitree G1 |
a7-cabinet-close/ |
Close the cabinet door | Unitree G1 |
a18-pour-water/ |
Pour water from a pitcher into a glass | H2 |
Each task folder has epoch-010/ and epoch-020/ (100 demonstrations per task, 20 epochs, saved at epoch 10 and 20).
Every epoch folder is self-contained: sharded safetensors weights, config.json, processor config and normalization
statistics, the full training config (experiment_cfg/), the dataset modality layout (dataset_meta/) and
protocol.json (dimensions, step counts, file list). Only what evaluation needs is included (no optimizer / trainer / resume state).
The VLM backbone nvidia/Cosmos-Reason2-2B is loaded from its own (gated) repository.
code/ holds the adapter code used for training.
Training used the Humanoid Benchmark online data augmentation spec v2 (photometric, geometric, sensor and lighting augmentation plus proprioception noise/delay/drop), with GR00T's own image augmentation bypassed.