Downloads · 30 days
0
microduck-skills
torchrl/microduck-skills
microduck-skills is a reinforcement learning model from torchrl. Use it for the reinforcement learning task on the model card, and read the license before you ship it in a product. It is set up for torchrl.
This repository exposes one canonical nine-skill locomotion prior and one canonical export for each published game. Each artifact has a stable path, one inference checkpoint, one video, and the metadata needed to unde…
Access
Public
Updated Sep 18, 2026
Repo size
12.7 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.mp44.5 MB · 74%
At a glance
- Task
- Reinforcement Learning
- Library
- torchrl
- Access
- Public
- Created
- Sep 11, 2026
- Updated
- Sep 18, 2026
- SHA
- d6766234
Model card
From the Hugging Face model README
MicroDuck policies for TorchRL
This repository exposes one canonical nine-skill locomotion prior and one canonical export for each published game. Each artifact has a stable path, one inference checkpoint, one video, and the metadata needed to understand how it was produced.
| Artifact | Checkpoint | Video | Status |
|---|---|---|---|
| Nine-skill locomotion prior | prior/checkpoint.ckpt | prior/video.mp4 | Validated |
| Football selector | games/football/checkpoint.ckpt | games/football/video.mp4 | Experimental |
| Tag selector | games/tag/checkpoint.ckpt | games/tag/video.mp4 | Unsolved pilot |
The game selectors depend on the canonical locomotion prior; they do not contain or replace its weights. Read the artifact README before loading a checkpoint:
Nine-skill prior
The prior contains standing, forward and backward walking, left and right sidestepping, forward hopping, left and right turning, and in-place hopping. It is the selected checkpoint from the September 2026 training lineage and is the only locomotion checkpoint published at the repository's current head.
<video src="https://huggingface.co/torchrl/microduck-skills/resolve/main/prior/video.mp4" controls width="960"></video>
Its complete phased hyperparameters, initialization lineage, task ordering,
evaluation metrics, acceptance report, and hashes remain next to the
checkpoint. In particular, this was not a from-scratch nine-skill run: it
started from the earlier six-skill actor and then underwent four phases of
joint nine-skill training. See prior/training.json.
Versioning
The paths above always identify the currently selected artifacts. For reproducible use, pin the Hugging Face repository commit returned when you download them. Replaced and removed exports remain available in the repository history; they are intentionally not duplicated in the current file tree.
All policies were evaluated in simulation. Hardware deployment has not been validated.
- Task
- Reinforcement Learning
- Library
- torchrl
- Created
- Sep 11, 2026
- Updated
- Sep 18, 2026