Request sample access
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
This repository contains the dataset specification, annotation schema, and sample metadata files. The production dataset (6,000 hours of sensor-rich human manipulation demonstrations with synchronized video, RGB-D, mocap, IMU, and pose data) is rights-cleared and delivered directly under a commercial license. Approved requesters get the full annotation schema and data dictionary in this repository, and can request a review package with real multi-sensor samples and QA summaries. Requests are reviewed within 1 business day.
Log in or Sign Up to review the conditions and access this dataset content.
Multimodal Human Manipulation Dataset
6,000 hours of sensor-rich human manipulation demonstrations: egocentric and external video, RGB-D, motion capture, IMU, hand pose, object labels, task phases, and interaction annotations for Physical AI and robotics.
This repository contains the full technical specification, annotation schema, and sample metadata files (Parquet). The production dataset is rights-cleared and delivered directly to buyers. Request access to see the full schema and get real multi-sensor samples.
Overview
The Multimodal Human Manipulation Dataset is a 6,000-hour sensor-rich collection of humans performing physical manipulation tasks with synchronized visual, motion, and sensor data. Unlike robot teleoperation data, this dataset captures human demonstrations rather than robot executions, giving robotics and Physical AI teams richer supervision than video alone without requiring full robot action trajectories.
Models get a detailed view of how humans reach, grasp, transfer, fold, wipe, open, close, sort, place, and manipulate objects in real-world physical environments. The dataset supports physical reasoning, human demonstration learning, action recognition, motion modeling, manipulation research, embodied AI, and world model training for real-world object interaction.
At a glance
| Total volume | 6,000 hours |
| Demonstration type | Human (not robot teleoperation) |
| Camera coverage | First-person egocentric + third-person views where available |
| Sensor modalities | RGB-D / depth where available, motion capture, IMU, hand pose, body pose |
| Annotations | Object labels, tool labels, task phases, contact events, completion states |
Modalities
Each recording may include:
- First-person egocentric video
- Third-person camera views
- RGB-D video and depth data
- Motion capture
- IMU signals
- Hand pose and body pose
- Object labels, tool labels, timestamps, task phase labels, completion states
Task coverage
Pick-and-place, object sorting, drawer opening and closing, folding, wiping, bin loading and unloading, shelf stocking, household manipulation, tool use, light assembly, package handling, object transfer, and other contact-rich manipulation workflows.
Technical specifications
- Annotation coverage: task phase, action segment, object labels, tool labels, hand-object contact, completion state, before/after state
- Buyer evaluation metrics: sensor synchronization accuracy, hand pose coverage, object visibility, depth availability, task completion labels, multi-view alignment
- QA metrics: calibration checks, timestamp alignment checks, sensor dropout flags, video usability score, annotation quality score
- Delivery format: synchronized video, sensor logs, CSV metadata, JSON annotations, pose files where available
Annotation schema
The gated file annotation_schema.json in this repository contains the full session schema with an illustrative example record, including per-stream sensor descriptors, sync and calibration QA fields, and contact-event annotations.
The samples/ folder holds sample metadata in Parquet format: index_sample.parquet (item-level) and events_sample.parquet (event-level), both conforming to this schema. Values are generated to illustrate structure and field distributions; production records ship in buyer review packages.
How to evaluate this dataset
- Request access using the form above. Requests are reviewed within 1 business day.
- On approval you get the gated files in this repository: full annotation schema, sample metadata Parquet files, data dictionary, and access instructions.
- Request a review package and we deliver real synchronized multi-sensor samples in your target task categories, sensor logs, JSON annotations, QA summaries, and licensing documentation within 2 business days.
All samples are delivered with structured CSV metadata and JSON annotation files where available. Buyer review packages include representative media files, metadata samples, annotation schema, QA summaries, and data dictionary documentation.
Licensing
The production dataset is rights-cleared for commercial AI training and licensed directly by Datoric, with documented contributor consent and chain-of-custody. Subset (by modality, task category, or hours), exclusive, and custom-collection options are available.
About Datoric
Datoric supplies rights-cleared, spec-exact training data for frontier AI labs and enterprise model teams: human manipulation data, robot episodes, egocentric and industrial video, expressive multilingual voice, computer-use traces, and gameplay trajectories. We also run managed collection pipelines for custom sensor stacks and specifications.
Contact: nikhil@arzule.com
- Downloads last month
- -