Downloads · 30 days
4
16% of all-time downloads
zys1030/smolvla-block-rot-sa-70ep-50k
smolvla-block-rot-sa-70ep-50k is a robotics model from zys1030. 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 lerobot. The card lists the license as apache-2.0.
This repository contains an expert-only SmolVLA checkpoint trained on the 70-episode shared-autonomy-assisted dataset:
Downloads · 30 days
4
16% of all-time downloads
All-time downloads
25
Public
Parameters
450M
907 MB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors907 MB · 100%
How the weights are stored.
BF16447M · 99%
From the Hugging Face model README
This repository contains an expert-only SmolVLA checkpoint trained on the 70-episode shared-autonomy-assisted dataset:
zys1030/sharedautonomy-vla-block-rot-sa-70ep
| Property | Value |
|---|---|
| Policy family | SmolVLA |
| Adaptation | Expert-only |
| Training steps | 50,000 |
| Training data | 70 episodes, 14,212 frames |
| Observations | Two RGB video streams plus a 10-D state vector |
| Actions | 7-D robot action vector |
| Action chunk size | 50 |
| Base policy | lerobot/smolvla_base |
| Vision-language model | HuggingFaceTB/SmolVLM2-500M-Video-Instruct |
The checkpoint contains the model weights and policy pre/post-processing artifacts required for inference. Optimizer, scheduler, and random-state training state are not included; this release is not intended for exact mid-run training resumption.
These are the final logged training values, not validation metrics:
| Metric | Value |
|---|---|
| Train loss | 0.013 |
| Gradient norm | 0.466 |
| Learning rate | 2.5e-6 |
No validation split was used for these training logs. Comparative training curves should be read together with the project's training documentation.
from lerobot.policies.smolvla.modeling_smolvla import SmolVLAPolicy
policy = SmolVLAPolicy.from_pretrained(
"zys1030/smolvla-block-rot-sa-70ep-50k"
)
Use a LeRobot environment with the SmolVLA dependencies installed. Physical robot execution must remain behind an independent safety supervisor and should not be enabled solely because the checkpoint loads successfully.
This policy was trained for a fixed laboratory pick-and-place task:
Pick up the red cube and place it in the UP region.
The checkpoint should not be treated as a general-purpose robot policy. Its behavior may be sensitive to camera placement, scene layout, object appearance, language phrasing, robot calibration, and distribution shift.
zys1030/sharedautonomy-vla-block-rot-manual-70epzys1030/smolvla-block-rot-manual-70ep-50kThis checkpoint is released under the Apache License 2.0. The checkpoint is
fine-tuned from lerobot/smolvla_base
using the LeRobot SmolVLA implementation and the
HuggingFaceTB/SmolVLM2-500M-Video-Instruct
vision-language model. Please retain the upstream attribution and license
notices when redistributing this checkpoint. See LICENSE and
UPSTREAM_ATTRIBUTION.md.
The training dataset is published separately under CC BY 4.0 with the
attribution Yun-Sheng Zhao, SharedAutonomy-VLA, 2026. The dataset license
is separate from this checkpoint's license; downstream users should review
the dataset card
for its terms and attribution requirements.