README.md

metadata

license: cc-by-nc-4.0
task_categories:
  - robotics
  - reinforcement-learning
  - depth-estimation
tags:
  - robotics
  - manipulation
  - embodied-ai
  - imitation-learning
  - rgbd
  - realsense
  - kitchen
  - cooking
  - human-demonstration

🍳 Chinese Commercial Kitchen Manipulation Dataset — Sample Pack v0.1

Asia's first real commercial kitchen manipulation dataset.
Professional chef (20 years) · Real restaurant environment · Multi-view RGB-D · Egocentric video

📧 Request evaluation samples or full data: andy@dynamicnova.com

---

Overview

This sample pack contains real-world cooking demonstrations collected in an operating Chinese commercial kitchen in Zhongshan, Guangdong, China.

The data focuses on professional chef workflows rather than staged tabletop demonstrations. It includes synchronized or task-aligned multi-view video, egocentric footage, and metric depth data for evaluating robotics, embodied AI, imitation learning, and visual action understanding pipelines.

Related work such as EgoMimic suggests that egocentric human demonstration data can be valuable for scaling imitation learning, especially when paired with robot data or aligned sensing setups.

Key characteristics:

• Real commercial Chinese restaurant kitchen
• Professional chef with approximately 20 years of experience
• Egocentric, side-view, and overhead camera perspectives
• Intel RealSense D435I RGB-D capture for overhead view
• Chinese cooking tasks involving tool use, bimanual coordination, and fine-grained food manipulation

Note: Large egocentric videos in this sample pack are provided as 1080p preview encodings for easier download. Full 4K source recordings are available upon request.

---

Sample Pack Contents

Task 1 — Cutting Vegetables (切菜) · 3 camera views + depth

File	Description
egocentric_1080p.mp4	Head-mounted action camera, first-person view (1920×1080, 30fps preview)
side_view.mp4	Fixed side-view phone camera (1920×1080, 60fps, original capture)
overhead/overhead.mp4	Fixed overhead RealSense RGB (1280×720, 15fps)
depth/depth.hdf5	Aligned depth frames, float32 in meters (480×848, 2379 frames)

Task 2 — Wok Stir-Fry (翻炒) · 2 camera views

File	Description
egocentric_1080p.mp4	Head-mounted action camera, first-person view (1920×1080, 30fps preview)
side_view.mp4	Fixed side-view phone camera (1920×1080, 30fps preview)

Depth sample is included for Task 1. Additional depth recordings may be available depending on the task and capture setup.

Depth note: Pixels with value 65.535m indicate no valid depth return (sensor limit). Typical valid pixel rate: ~86%.

Preview PNG/JPG images are included alongside the videos for quick browsing.

---

File Structure

task_01_cutting/
├── egocentric_1080p.mp4
├── side_view.mp4
├── overhead/
│   ├── overhead.mp4
│   └── overhead_*.PNG
├── depth/
│   ├── depth.hdf5
│   └── check_*.jpg
├── egocentric_screenshot_*.PNG
└── side_view_*.PNG
task_02_stir_fry/
├── egocentric_1080p.mp4
├── side_view.mp4
├── egocentric_screenshot_*.PNG
└── side_view_*.PNG

Read depth data (Python):

import h5py
with h5py.File("task_01_cutting/depth/depth.hdf5", "r") as f:
    depth = f["depth_meters"][:]   # (2379, 480, 848) float32, meters
    ts    = f["timestamps"][:]

---

Camera Setup

Sample pack (this repository)

View	Task 1	Task 2
Egocentric	1920×1080, 30fps (egocentric_1080p.mp4)	1920×1080, 30fps (egocentric_1080p.mp4)
Side	1920×1080, 60fps (side_view.mp4)	1920×1080, 30fps (side_view.mp4)
Overhead RGB	1280×720, 15fps	—
Overhead depth	848×480 (HDF5)	—

Original capture specs (full dataset on request)

View	Device	Resolution	Frame Rate
Egocentric	Head-mounted action camera (DJI Osmo Action 3)	3840×2160 (4K)	29.97 fps
Side	Fixed smartphone	1920×1080	up to 60 fps
Overhead RGB-D	Intel RealSense D435I	RGB 1280×720 / Depth 848×480	15 fps

---

Collection Environment

• Location: Zhongshan, Guangdong, China
• Venue: Operating commercial Chinese restaurant
• Operator: Professional chef, 20 years experience
• Consent: Full informed consent obtained from all participants

---

Tasks

Task	Chinese	Difficulty	Bimanual	Camera Views	Depth
Cutting vegetables	切菜	Medium	Partial	3	✅
Wok stir-fry	翻炒	High	✅	2	Available upon request

Planned in full dataset:

Task	Chinese	Key Challenge
Dumpling folding	包饺子	High dexterity, bimanual
Dough kneading	揉面	Force estimation, rhythm
Deep frying	炸	Temperature judgment, timing
Pan frying	煎	Heat control, single/double-side flip
Braising / stewing	炖煮	Long-horizon, multi-step sequencing
Sauce thickening	勾芡	Fine motor control, timing-sensitive
Marinating / seasoning	腌制/调味	Multi-ingredient coordination

Action-level labels (e.g. flip timing, thickening moment, seasoning sequence) available upon request.

---

Access to Full Data

This public sample pack is intended for technical evaluation and early research feedback.

Additional materials may be available upon request:

• Full 4K egocentric MP4 recordings
• Longer multi-view MP4 recordings
• Additional HDF5 metric depth sequences
• RealSense raw .bag recordings
• Task-level or action-level annotations
• Format conversion support, including HDF5, RLDS, or LeRobot
• Custom collection for specific kitchen workflows

---

Contact

📧 andy@dynamicnova.com

Please include: tasks of interest, required volume, preferred format (HDF5 / RLDS / LeRobot), timeline.

Response within 48 hours.

---

Citation

If you use this dataset in academic or commercial research, please cite this repository.

---

Collected May 2026 · Zhongshan, Guangdong, China · Nova Dynamics Limited