Downloads · 30 days
4
12% of all-time downloads
jren313/g1-inspire-piston-starvla-gr00t-n1d7
g1-inspire-piston-starvla-gr00t-n1d7 is a robotics model from jren313. Use it for the robotics task on the model card, and read the license before you ship it in a product. It is set up for starvla. The card lists the license as other.
A GR00T-N1.7 fine-tune for a Unitree G1 (29-DoF) with Inspire hands performing a piston pick-and-place task, trained inside starVLA with the CosmosGR00TN1d7 framework.
Downloads · 30 days
4
12% of all-time downloads
All-time downloads
33
Public
Repo size
9.5 GB
Likes
0
Public
Click a slice to open those files.
.pt9.5 GB · 100%
From the Hugging Face model README
A GR00T-N1.7 fine-tune for a Unitree G1 (29-DoF) with Inspire hands performing a
piston pick-and-place task, trained inside starVLA
with the CosmosGR00TN1d7 framework.
Unlike a from-scratch action head, this model is initialised directly from
nvidia/GR00T-N1.7-3B — all 1031 checkpoint tensors (VLM backbone and
flow-matching action head) load 1:1 — and then fine-tuned end-to-end on the
piston dataset.
| Component | Detail |
|---|---|
| Backbone | Cosmos-Reason2-2B (Qwen3-VL architecture), LLM truncated to select_layer=16 (16 of 28 layers) |
| Action head | 32-layer "alternate-VL" flow-matching DiT — every block self-attention + adaLN; even blocks cross-attend to VL features, alternating text-only / image-only tokens every 2 blocks |
| VL adapters | 4-layer VL self-attention transformer + VL LayerNorm (vlln) |
| Projectors | Multi-embodiment (category-specific) state encoder / action encoder / action decoder, 32 embodiment slots, 132-wide |
| Objective | Rectified flow matching (velocity prediction), 4 Euler denoising steps at inference |
There are no "future tokens"; the DiT contains no cross-attention blocks in the classic sense — VL conditioning enters via the alternating cross-attend pattern.
"pick up the piston."[left_arm 7 | right_arm 7 | left_hand 6 | right_hand 6 | waist 3][left_arm 7 | right_arm 7 | left_hand 6 | right_hand 6 | base_height 1 | navigate_command 3]25 — GR00T's pretrained
unitree_g1_full_body_with_waist_height_nav_cmd, whose action layout matches this
robot, so the category-specific projectors warm-start rather than starting cold.dataset_statistics.json).
Real 30-dim action / 29-dim state are zero-padded to the head's 132-wide projectors
at runtime; the loss is masked back to the real dims.birbirll/g1-inspire-piston-pick-place
(LeRobot v2.1, 102 episodes, 50 fps, single 240×424 ego view)tune_llm=true, tune_visual=true)1e-5, action head 1e-4, cosine schedule with min LR 1e-6, 50 warm-up stepsPagedAdamW8bit, betas (0.9, 0.95), wd 1e-8, grad clip 1.0state_dropout_prob = 0.8 — proprioceptive state is available but zeroed 80% of
the time during training, so the policy is strongly camera-driven and does not
lean on proprio. This matters for sim/real transfer: a state-reliant policy can
minimise training loss while ignoring the image, then fail to correct visually in
closed loop.accelerate (no DeepSpeed), gradient checkpointing, single GPU.config.yaml # resolved training config (required to load)
dataset_statistics.json # q99 norm stats (required to un-normalise actions)
final_model/pytorch_model.pt # trained weights (~8.9 GB)
starvla_gr00t_n1d7_piston.yaml # training config as authored
run_starvla_GR00T_N1d7_piston_train.sh # exact launcher used
config.full.yaml, summary.jsonl # provenance
config.yaml and dataset_statistics.json must sit one level above the
checkpoint file — starVLA's loader resolves them from the run root.
Requires starVLA with the CosmosGR00TN1d7 framework, plus nvidia/Cosmos-Reason2-2B
available locally (it is a gated repo — request access first).
Serve it over the GR00T ZMQ protocol:
python deployment/model_server/server_policy_gr00t_zmq.py \
--ckpt_path <run_dir>/final_model/pytorch_model.pt \
--port 5555 --use_bf16 --no_state \
--fallback_instruction "pick up the piston."
--no_state serves it camera-only, which is well matched to the 0.8 state dropout and
avoids depending on the client's proprio format. Drop the flag to feed real state.
For closed-loop control, a GR00T-protocol client (e.g. a WBC bridge) should use the UNITREE_G1 contract — absolute joints, single ego frame, 30-step chunks — not the "REAL_G1 / N1.7" contract (relative arms, 2-frame video, 40-step), despite this model's N1.7 architecture. The protocol and the architecture are independent choices here.
Derived from NVIDIA GR00T-N1.7 (nvidia/GR00T-N1.7-3B) and
NVIDIA Cosmos-Reason2-2B, and contains fine-tuned weights of both. The DiT and
embodiment-conditioned MLP modules in starVLA are vendored from
Isaac-GR00T (Apache-2.0). Use of these weights
is subject to NVIDIA's licence and model-access terms for the upstream models in
addition to anything stated here.