Kyry10/k1-dances

K1 Dances

Booster K1 dance dataset: 1046 dances (3.1 hours) from the AIST Dance Video Database, retargeted to the 22-DoF K1. All 1046 are trackable zero-shot by SONIC-K1.

Website · Space · SONIC-K1 · Required citation · AIST Dance Video Database

These dances belong to AIST. Every motion here comes from the AIST Dance Video Database, created by AIST (National Institute of Advanced Industrial Science and Technology, Japan), through the AIST++ 3D annotations by Google. We only retargeted the dancers' movements to a robot. If you use this data, you must cite Tsuchida et al., ISMIR 2019. Commercial use is subject to the AIST Dance DB Terms of Use.

At a glance

Dances	1046
Styles	10
Total length	187.0 minutes at 60 fps
Robot	Booster K1, 22 DoF
Dance source	AIST Dance Video Database / AIST++ (arXiv:2101.08779)
Retargeting	UMR (arXiv:2609.02134), K1 config
Trackable by SONIC-K1	all 1046, zero-shot: SONIC-K1 v1.0 (arXiv:2610.39327), the K1 port of NVIDIA SONIC, MuJoCo simulation, no fine-tuning
Pass rate	1046 of 1363 AIST++ clips (76.7%), median tracking error 24.3 mm

Styles

House · 121	Break · 62	Waack · 87
Krump · 116	Pop · 136	Lock · 130
LA-style Hip-hop · 126	Middle Hip-hop · 121	Street Jazz · 115

Plus Ballet Jazz · 32. Numbers are dances per style. More videos on the website.

Use it for

• Motion-imitation RL on the Booster K1: whole-body tracking targets for BeyondMimic-style trackers, mjlab, Isaac Lab or MuJoCo (convert qpos to your trainer's motion format).
• Playing dances with SONIC-K1 without training: every clip here is already trackable by it.
• Testing whole-body controllers on diverse, fast, expressive motion.
• Robot dance and animation: 10 styles from House and Break to Krump and Waack.
• Retargeting research: every clip keeps its AIST ID, so you can compare against the original AIST++ motion.

Quick start

import numpy as np
with np.load("motions/gBR_sBM_cAll_d04_mBR0_ch01.npz", allow_pickle=False) as m:
    qpos = m["qpos"]            # T x 29
    fps = float(m["fps"])       # 60
root_xyz  = qpos[:, :3]         # metres, world Z-up
root_wxyz = qpos[:, 3:7]        # unit quaternion, WXYZ
joints    = qpos[:, 7:]         # 22 joints, radians

Joint order is in joint_names.json and in each file's joint_names (the SONIC-K1 k1.xml order). Quick check: python load_motion.py motions/<id>.npz. These are motion references (targets for a tracking controller), not low-level robot commands.

What's inside

File	Contents
motions/<id>.npz	Retargeted reference: qpos (T x 29, float32), fps, joint_names, method, info
previews/	Animated 3 s preview of every dance + metadata.csv; this is the table in Data Studio
manifest.csv	One row per dance: style, routine, dancer and music IDs, duration, SONIC tracking metrics, quality flags, SHA-256
joint_names.json	The 29 qpos columns and joint limits
summary.json	Counts and durations
quality_criteria.json	Definition of the stricter quality screen
CITATION.bib, LICENSE.md	Required citation and use conditions
assets/	Preview video and images for this page
SHA256SUMS	Integrity check: shasum -a 256 -c SHA256SUMS

IDs keep the AIST naming, e.g. gBR_sBM_cAll_d04_mBR0_ch01 = genre Break, basic routine, dancer 04, music mBR0, choreography 01.

How the dances were selected

1. All 1363 usable AIST++ clips (the official list minus the AIST++ ignore list) were retargeted to the K1 with UMR, at the original timing.
2. Each clip was run through the released SONIC-K1 v1.0 controller in full-clip MuJoCo simulation, no fine-tuning, 0 ms action delay. A clip passes if the robot completes it without falling (root height error ≤ 0.25 m, orientation error ≤ 1 rad) with mean local tracking error ≤ 50 mm.
3. All 1046 passing clips are included. The sonic_* columns give per-clip tracking quality.
4. strict_quality_screen = True marks 391 clips that also pass a stricter check for self-collision, floor penetration, foot sliding and joint speed (quality_criteria.json).

Limitations

• Simulation only. These dances were not run on a real K1. SONIC-K1 v1.0 itself runs on a physical K1, but dances are fast and real-robot results may be much lower.
• Tracking was checked with 0 ms action delay only.
• Some clips contain brief self-collision, foot sliding or a fast arm flip (a joint swinging between its limits within a few frames, see ref_max_joint_speed_radps). The checks are automated, not a frame-by-frame review.

Required citation and use conditions

The dances are AIST's. This dataset is released under the conditions set by the AIST Dance DB team:

1. Everyone who uses or redistributes this data must cite Tsuchida et al., AIST Dance Video Database, ISMIR 2019, and keep this requirement when redistributing.
2. Commercial use is subject to the AIST Dance DB Terms of Use.
3. No original AIST content (video, music, AIST++ SMPL annotations) is included or licensed here.

@inproceedings{tsuchida2019aist,
  title={AIST Dance Video Database: Multi-genre, Multi-dancer, and Multi-camera Database for Dance Information Processing},
  author={Tsuchida, Shuhei and Fukayama, Satoru and Hamasaki, Masahiro and Goto, Masataka},
  booktitle={Proceedings of the 20th International Society for Music Information Retrieval Conference (ISMIR)},
  year={2019}
}

Please also cite AIST++, UMR, SONIC-K1 and SONIC (all entries are in CITATION.bib), and this dataset:

@misc{kolesnichenko2026k1dances,
  title={K1 Dances: AIST++ Dances Retargeted to the Booster K1 Humanoid},
  author={Kolesnichenko, Kyrylo},
  year={2026},
  howpublished={\url{https://huggingface.co/datasets/Kyry10/k1-dances}}
}

Credits

Dances: AIST Dance Video Database (AIST) and AIST++ (Google). Retargeting: UMR. Tracking check: SONIC-K1, built on NVIDIA SONIC. Robot model: Booster K1. Made by Kyrylo Kolesnichenko, ATR Lab, Kent State University. Same dances for the Unitree G1: G1 Dances.