Instructions to use djduecaster/dfreq-talent with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Microduck
How to use djduecaster/dfreq-talent with Microduck:
sudo robotctl policy load walk djduecaster/dfreq-talent
- Notebooks
- Google Colab
- Kaggle
|
Download README.md from djduecaster/dfreq-talent: direct link, hf CLI and curl.
- Browser
- Download file 3.14 kB
-
https://huggingface.co/djduecaster/dfreq-talent/resolve/main/README.md
- Command line
-
hf download hf://djduecaster/dfreq-talent/README.md
-
curl -L -o README.md https://huggingface.co/djduecaster/dfreq-talent/resolve/main/README.md
3.14 kB
| tags: | |
| - microduck | |
| - robotics | |
| - onnx | |
| - microduck-slot:walk | |
| library_name: microduck | |
| pipeline_tag: robotics | |
| # Seven Nation Quack — five-duck performance | |
| A 20-second Microduck Arena Talent performance: one vocal lead, two riff voices and two beat voices perform a short synthesized-vowel medley. The lead makes a walking entrance; the four backing ducks perform beat-timed step accents and short mirrored turns. No commercial recording or intelligible lyrics are used. | |
| ## Files and commands | |
| `policy.onnx` is unchanged Pollen Robotics `BEST_alpha_walking` feedback weights. `timeline.json` is a version-2 cast of five `self` ducks sharing that policy; it contains the complete music, placement and movement commands. No additional borrowed model repository is needed. For this walking controller, `twist` means forward speed (m/s), lateral speed (m/s), and yaw rate (rad/s). Head/body commands are zero. This is not the Talent sit/stand controller. | |
| The Arena renders the timeline's native vowel voice and movement. Local review-video camera cuts, explanatory captions, role colors and the tiny delivery fade are not timeline features, and the MP4/WAV are not the public performance payload. | |
| ## Attribution and reproducibility | |
| Pollen Robotics weights are reused unchanged from the official Arena repository: | |
| https://github.com/pollen-robotics/microduck-arena/blob/bbddfe587d300063b1d162281d647740738742f5/arena/assets/policies/BEST_alpha_walking.onnx | |
| Policy SHA-256: `e36332d383997d51401897734cd3e79cf5038406feddb18b4d57ecfb141daa6c`. No new neural training, weight editing or custom export was performed. The contribution here is the external five-duck command timeline and short vowel arrangement. No standalone open-source license for the inherited neural weights is asserted. | |
| Music reference: the authorized public piano/vocal/guitar preview for The White Stripes' *Seven Nation Army*, Musicnotes edition MN0111227 (Universal Music Publishing Group): | |
| https://www.musicnotes.com/sheetmusic/the-white-stripes/seven-nation-army/MN0111227 | |
| The short arrangement selects source-grounded motifs into an edited medley; it is not a full-song recording or a claim of composition-rights clearance. Native vowels approximate pitch and syllable rhythm, not words. No full lyrics or score is distributed. | |
| ## Local evidence | |
| Before preparation, three private full-cast 20-second rehearsals with Arena runner 5 (seeds 1, 1, 101) finished with no falls or interduck contacts. Seed 1 trace/settle arrays matched exactly. All collision geometry stayed inside the stage. The lead entrance advanced about 15–21 cm across the tested starts; measured foot sole clearance reached about 1.78 cm. Beat commands share a 120 BPM grid; exact learned-gait footfall phases vary, so perfectly synchronized heavy stomps are not claimed. No one-foot spins or kinematic foot animation. | |
| These are local simulation checks, not a server result or real-robot deployment claim. The active public Talent performance changes only after a new Arena entry plays to completion. Votes remain attached to their original run and do not transfer to replacements. | |