Datasets / Physical AI

Robot Teleoperation Dataset

Custom collection · specified robot

Demonstrations recorded on a specified robot, with aligned video, states, actions, and documented sensors.

Request access on Hugging Face

Dataset specification

  • Collected on demand on buyer-specified robot platforms and tasks
  • Reference rig: three synchronized stereo RGB streams at 1280×720 and 15 fps
  • Stereo-derived depth, with camera and sensor configuration agreed before collection
  • Robot states and actions at platform-native control rates
  • Robot platform, action-space convention, gripper encoding, and camera calibration documented
  • RLDS/TFRecord, LeRobot, HDF5, ROS bag, or a custom delivery schema

Recorded content

  • Demonstrations on a specified robot
  • Aligned video, states, and actions
  • Documented robot and sensor configuration

Scope notes

  • This is custom collection. The number of recordings, robots, tasks, and schedule are agreed before starting.
  • The reference camera rig is configurable and is not a fixed requirement for every collection.
  • State and action rates depend on the selected robot platform.

Collection scope

Custom collection

  • Custom collection on buyer-specified platforms and tasks
  • Number of recordings agreed before collection

Reference capture rig

  • Three synchronized stereo RGB streams at 1280×720 and 15 fps
  • Stereo-derived depth; camera count, resolution, and sensors agreed before collection

Robot interface

  • Robot states and actions at platform-native control rates
  • Documented action-space conventions, gripper encoding, and camera extrinsics

Delivery

  • Agreed file format, including Open X-Embodiment-compatible RLDS and LeRobot
  • HDF5, ROS bag, or a custom schema with JSON metadata

Annotation & metadata fields

  • Task and object labels
  • Task phases
  • Success and failure
  • Retry and recovery labels
  • Robot platform and action-space convention
  • Gripper encoding, camera extrinsics, and calibration

Capture methodology

  • Consented operators teleoperate the specified robot through task episodes
  • Multi-view stereo video, derived depth, robot states, and actions are recorded together
  • Camera and sensor configuration is agreed before collection
  • The delivery includes documentation of the robot, sensors, and controls

Provenance & rights chain

  • Recorded by consented participants or operators
  • Licensing and the selected delivery scope are reviewed with Datoric

Quality, duplicates & PII

How submissions are reviewed and cleaned before they are accepted into the dataset.

  • Timestamp and stream-alignment checks
  • Dropped-frame checks
  • State and action log completeness
  • Outcome review

Formats & delivery

  • Open X-Embodiment-compatible RLDS/TFRecord
  • LeRobot
  • HDF5 or ROS bag
  • Custom schema with JSON metadata and robot configuration records

Rights & license scope

Licensed by Datoric for robotics training, with final scope governed by the agreement for the selected delivery.

Frequently asked

Is this a fixed inventory of robot episodes?
No. New data is recorded on a specified robot for the agreed tasks. The number of recordings and delivery dates are set before collection.
Can you deliver LeRobot or RLDS?
Yes. The catalog includes LeRobot and Open X-Embodiment-compatible RLDS/TFRecord, alongside HDF5, ROS bag, and custom schemas.
What robot setup is documented?
The delivery documents the robot platform, action-space convention, gripper encoding, camera extrinsics, and calibration so the recorded data can be interpreted correctly.

Related dataset specifications