Skip to content
CHOSHU·DATA
2026-11-09 · 6 min read

Japanese Datasets for Robotics

Environments, objects, and etiquette that embodied models have to learn locally.

Embodied models generalize across embodiments better than they generalize across environments. Japan offers distinct ones.

Spaces

Compact apartments, genkan entryways, tatami rooms and small kitchens change reachability and navigation planning. Industrial settings have their own conventions for safety and process.

Objects

Rice cookers, chopsticks, futon and local packaging sit outside typical household-object datasets.

Data types

  • Egocentric video with task segmentation
  • Teleoperated demonstrations with proprioception and force
  • 3D scans of objects and rooms
  • Human motion capture

Consent is the hard part

Home data involves household members, neighbors and visible documents. Collection should use paid, consented participants, on-device or post-capture anonymization and a documented review step. Factory footage requires explicit site-owner rights.

General information, not legal advice.

Next
Why Translated Preference Data Isn't Enough for Japanese LLMs
Next
Provenance Is Becoming a Buying Criterion for Japanese Training Data
Start here

Tell us what your model is missing.

Describe the data you need. We will respond with feasibility, rights approach and next steps.