LiveNew tools this week

gr00t_n1_7_openarm_wam_v1_50k

seungkukim/gr00t_n1_7_openarm_wam_v1_50k

gr00t_n1_7_openarm_wam_v1_50k is a robotics model from seungkukim. 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 gr00t. The card lists the license as apache-2.0.

Fine-tune of nvidia/GR00T-N1.7-3B on the six robot subsets of openarmwamv1, a 28-dim bimanual OpenArm embodiment with two ego cameras.

Downloads · 30 days

14

39% of all-time downloads

All-time downloads

36

Public

Parameters

3.1B

6.9 GB on disk

Likes

0

Public

Repo makeup

Click a slice to open those files.

.safetensors6.9 GB · 100%

At a glance

Task: Robotics
Library: gr00t
License: apache-2.0
Model type: Gr00tN1d7
Access: Public
Created: Sep 19, 2026
Updated: Sep 19, 2026
SHA: 46d75542

Model card

From the Hugging Face model README

GR00T N1.7 — OpenArm WAM v1 (50k steps)

Fine-tune of nvidia/GR00T-N1.7-3B on the six robot subsets of openarm_wam_v1, a 28-dim bimanual OpenArm embodiment with two ego cameras.

Training data

Subset	Episodes	Frames	fps
`openarm_ego_jungwook`	151	58,023	30
`openarm_teleop_v3/bottle`	96	35,506	20
`openarm_teleop_v3/cup`	108	28,757	20
`openarm_teleop_v3/doll`	48	13,414	20
`openarm_teleop_v3/snack`	96	40,184	20
`banana_v21_openarm28`	54	16,202	30
total	553	192,086

Mixture weights are frame-proportional (len_i / sum(len)).

Configuration

Embodiment tag: NEW_EMBODIMENT, 28-dim state/action
Action horizon 24; arms relative, neck + hands absolute
q01/q99 percentile normalization (--use-percentiles)
50,000 steps, global batch 32, LR 1e-4, cosine decay, warmup ratio 0.05, weight decay 1e-5
State dropout 0.2; video view dropout 0.5 (50% both views / 25% left / 25% right)
Images kept at native 192×256 (--shortest-image-edge 192 --crop-fraction 0.95)
1.62 B of 3.14 B parameters trainable (51.5%)

Trained on 2×H100 80GB in 3 h 41 m. Final training loss 0.0195.

Usage

from gr00t.policy.gr00t_policy import Gr00tPolicy

policy = Gr00tPolicy(
    model_path="seungkukim/gr00t_n1_7_openarm_wam_v1_50k",
    embodiment_tag="new_embodiment",
)
action = policy.get_action(observation)

Or serve it:

python gr00t/eval/run_gr00t_server.py \
    --model-path seungkukim/gr00t_n1_7_openarm_wam_v1_50k \
    --embodiment-tag new_embodiment

Contents

Inference-ready weights, processor and configs. The DeepSpeed optimizer state is not included, so this repo cannot resume training.

Task: Robotics
Library: gr00t
Type: Gr00tN1d7
License: apache-2.0
Created: Sep 19, 2026
Updated: Sep 19, 2026