You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

Dataset Card for Amplibotics Samples

Tasks Included in This Sample

Laundry Manufacturing Kitting Waste Disposal
  • pants_folding
  • parts_unbagging
  • tool_kitting
  • waste_sorting

Overview

Sample raw capture data from teleoperated bimanual demonstrations. Contact us at info@amplibotics.ai to request the full datasets, or to inquire about custom data solutions.

  • Created by: Amplibotics
  • Recorded at: Amplibotics Robot Farm

Our data collection technology works anywhere in the world, scaling up datasets with our network of remote robot pilots. Want data recorded from a particular location, on specific hardware, or doing custom tasks? Contact us to eliminate operational overhead and lead times for your AI projects.

Are you working on deploying robotics AI to real-world jobsites? Our real-time supervision & intervention service provides reliability guarantees — while mining high-value corrective datasets from operator actions. Amplibotics turns work environments into self-sustaining data flywheels, unlocking practical and cost-effective automation on day 1. Learn more at https://www.amplibotics.ai.

Environments

  • tabletop work surfaces (with task-specific materials & objects)

Embodiments

  • bimanual UFactory Lite6 arms configured with Amplibotics Parallel Jaw Grippers

Camera Configurations

  • overhead stereo pair (2x RGB) + 2x wrist (one per arm) | 1920x1200 at 30 FPS | hardware-synchronized global shutters

Control Methods

  • GELLO leader arms
  • flight stick (Thrustmaster) game controllers

Operators

  • Amplibotics pilots (trained for each task)

Directory Structure

.
├── episodes/
│   ├── pants_folding
│   │   └── <episode MCAPs>
│   ├── parts_unbagging
│   │   └── <episode MCAPs>
│   ├── tool_kitting
│   │   └── <episode MCAPs>
│   └── waste_disposal
│       └── <episode MCAPs>
├── foxglove_layouts/
│   ├── TaskLayout.json
│   └── DataLayout.json
└── README.md

File Type

MCAP

Metadata Records

Record Description Count Schema
session Session-level metadata including equipment configuration and identifiers. Exactly 1 session.md
episode_<uuid> Episode-level metadata for each recorded task attempt. At least 1 episode.md

Topics

Topic Pattern Description Schema Encoding Count
/robot_<name>/state Robot arm joint states end-effector pose. RobotState protobuf 1-2
/robot_<name>/command Commanded end-effector pose and velocity targets. RobotCommand protobuf 1-2
/robot_<name>/dh_parameters Denavit-Hartenberg kinematic parameters for the robot arm. RobotDHParameters protobuf 1-2
/tool_<name>/state Gripper state. ToolState protobuf 1-2
/tool_<name>/command Commanded gripper position. ToolCommand protobuf 1-2
/tool_<name>/info Tool info. GripperInfo protobuf At least 1
/camera_<name>/image Compressed video frames from cameras. CompressedVideo protobuf At least 1
/camera_<name>/info Camera intrinsic and extrinsic calibration parameters. CameraCalibration protobuf At least 1
/camera_<name>/extrinsics Camera extrinsic calibration parameters. HandEyeCalibration protobuf Any number
/tf_static Static coordinate frame transforms (e.g., camera mounts). FrameTransforms protobuf Exactly 1
/tf Non-static instantaneous TF trees. FrameTransforms protobuf Exactly 1
/episode/events Episode lifecycle events (start, end, subtasks). EpisodeEvent protobuf Exactly 1

session Metadata Record

Session-level metadata for mcap.bus.* bimanual teleoperation recordings.

Fields

Key Type Required Description
session-uuid string Yes Unique identifier for this capture session (UUID v4).
station-uuid string Yes Identifier for the physical workstation (UUID v4).
pilot-uuid string Yes Identifier for the pilot/operator who performed the session (UUID v4).
start-time-unix float Yes Session start time as Unix timestamp (seconds since epoch).
controller-type string Yes Type of controller hardware used for the session.
robot-left-model string Yes Model identifier for the left robot arm.
robot-right-model string Yes Model identifier for the right robot arm.
gripper-left-model string Yes Model identifier for the left gripper/end-effector.
gripper-right-model string Yes Model identifier for the right gripper/end-effector.
urdf-permalink string Yes URL to the URDF file describing the robot configuration.
collector-name string Yes Name of the data collection software.
collector-version string Yes Semantic version of the collector software.
schema-version string No data schema used by the collector.
data-contract string No Contract ID this file conforms to.
source-contract string No Contract ID of the source file this was transformed from.
episode-count integer No Number of episode metadata records.
axis-labels string No Axis labels.
button-labels string No Button labels.
job-id string No Job id.

episode_<uuid> Metadata Record

Episode-level metadata for mcap.bus.* bimanual teleoperation recordings.

Fields

Key Type Required Description
start_time integer Yes Episode start time in nanoseconds since Unix epoch.
end_time integer Yes Episode end time in nanoseconds since Unix epoch.
task string Yes Task instruction for this episode.
success boolean Yes Whether the episode completed successfully.
error string No Error message if the episode failed.
is_discard boolean No Whether this episode was marked for discard during collection.

Message Definitions

amplibotics/capture/capture.proto

Episode

A task completion attempt

Field Type Label Description
id string UUID: "123e4567-e89b-12d3-a456-426614174000"
instruction string Natural language task instruction

EpisodeEvent

Used for both episode lifecycle events and subtask events:

Field Type Label Description
timestamp google.protobuf.Timestamp protobuf Timestamp (int seconds & nanos)
event_type EpisodeEvent.EventType classification of this event
episode Episode set for EPISODE_* events & SUBTASK_* events
subtask Subtask only set for SUBTASK_* events
result TaskResult only set for *_END events

GripperCommand

Command sent to parallel-jaw gripper

Field Type Label Description
position float 0 is open, 1 is closed until effort threshold is met. Stored as float, but sent as boolean value for this gripper configuration.

GripperInfo

Parallel jaw gripper information

Field Type Label Description
open_span_mm float Gripper span when fully open (mm)
closed_span_mm float Gripper span when fully closed (mm)

GripperState

State of parallel-jaw gripper

Field Type Label Description
position float 'spans' travelled. 0 is open, 1 is closed
velocity float 'spans' per second
acceleration float 'spans' per second^2'
effort float Dynamixel current in mA

HandEyeCalibration

Field Type Label Description
calibrations HandEyeCalibrationEntry repeated One entry per robot this camera is calibrated to

HandEyeCalibrationEntry

Field Type Label Description
robot_position string Robot position name (e.g., "left")
calibration_type HandEyeCalibrationType
transform Pose Camera-to-reference transform from cv2.calibrateHandEye(): eye_in_hand: camera origin expressed in end-effector (gripper) frame eye_to_hand: camera origin expressed in robot base frame Translation in mm, orientation as unit quaternion.
reprojection_error float AX=XB translational consistency error (mm)
calibrated_at google.protobuf.Timestamp When calibration was performed

JointDHParam

DH parameters for a single joint

Field Type Label Description
d float Link offset in mm
a float Link length in mm
alpha float Link twist in radians
theta float Joint angle offset in radians

Pose

Pose given as xyz position and xyzw quaternion

Field Type Label Description
x float mm relative to robot's base frame
y float mm relative to robot's base frame
z float mm relative to robot's base frame
qx float Quaternion orientation (x) relative to robot's base frame
qy float Quaternion orientation (y) relative to robot's base frame
qz float Quaternion orientation (z) relative to robot's base frame
qw float Quaternion orientation (w) relative to robot's base frame

RawInput

Raw controller input signals

Field Type Label Description
axes float repeated Raw axis values
buttons bool repeated Raw button states

RobotDHParameters

DH parameters for a robot

Field Type Label Description
timestamp google.protobuf.Timestamp protobuf Timestamp (int seconds & nanos)
dh_parameters JointDHParam repeated DH parameters from robot
dh_tool_parameters JointDHParam repeated DH parameters with TCP offset applied
base_transform Pose Base-to-world mounting transform (identity = flat table mount)

RobotState

State reported by robot

Field Type Label Description
timestamp google.protobuf.Timestamp protobuf Timestamp (int seconds & nanos)
joint_positions float repeated rad
joint_velocities float repeated rad/s
joint_accelerations float repeated rad/s^2
joint_efforts float repeated estimated joint torques computed by robot's controller (Nm)
joint_position_sub_commands float repeated the applied joint position commands the robot’s controller is currently executing (rad)
ee_pose Pose end effector pose

Subtask

A subtask representing a shorter-horizon segment of the overall task

Field Type Label Description
instruction string Natural language subtask instruction

TaskResult

Information about the outcome of a task or subtask attempt

Field Type Label Description
success bool True if task completed successfully, false if not
label string 'success', 'failure', 'recovery', or 'impossible'

ToolCommand

Command sent to robot's tool/end-effector

Field Type Label Description
timestamp google.protobuf.Timestamp protobuf Timestamp (int seconds & nanos)
gripper GripperCommand

ToolState

State of robot's tool/end-effector

Field Type Label Description
timestamp google.protobuf.Timestamp protobuf Timestamp (int seconds & nanos)
gripper GripperState

EpisodeEvent.EventType

EpisodeEvent classification (enum)

Name Number Description
EVENT_TYPE_UNSPECIFIED 0
EPISODE_START 1 Event marks start of episode
EPISODE_END 2 Event marks end of episode
SUBTASK_START 3 Event marks start of subtask
SUBTASK_END 4 Event marks end of subtask

HandEyeCalibrationType

Name Number Description
HAND_EYE_CALIBRATION_TYPE_UNSPECIFIED 0
HAND_EYE_CALIBRATION_TYPE_EYE_IN_HAND 1 Camera mounted on robot wrist (camera-to-end-effector)
HAND_EYE_CALIBRATION_TYPE_EYE_TO_HAND 2 Camera fixed in world (camera-to-robot-base)

amplibotics/bus/bus.proto

RobotCommand

Extends capture RobotCommand with target_joint_velocities and target_joint_accelerations.

Field Type Label Description
timestamp google.protobuf.Timestamp protobuf Timestamp (int seconds & nanos)
target_ee_pose amplibotics.capture.Pose Absolute target end-effector pose (xyz in mm, orientation in quaternion). Commanded with leader arms, computed when using flight sticks.
ee_velocity amplibotics.capture.Pose End-effector velocity (xyz in mm/s, quaternion in 1/s). Commanded with flight sticks, computed when using leader arms.
target_joint_positions float repeated Target joint angles from IK (radians)
raw_input amplibotics.capture.RawInput Optional raw controller input
target_joint_velocities float repeated rad/s, computed from target_joint_positions
target_joint_accelerations float repeated rad/s², computed from target_joint_velocities

foxglove/foxglove.proto

CameraCalibration

Camera calibration parameters

Field Type Label Description
timestamp google.protobuf.Timestamp protobuf Timestamp (int seconds & nanos)
frame_id string
width fixed32
height fixed32
distortion_model string
D double repeated
K double repeated length 9
R double repeated length 9
P double repeated length 12

CompressedVideo

A single frame of a compressed video bitstream

Field Type Label Description
timestamp google.protobuf.Timestamp Timestamp of video frame
frame_id string Frame of reference for the video. The origin of the frame is the optical center of the camera. +x points to the right in the video, +y points down, and +z points into the plane of the video.
data bytes Compressed video frame data. For packet-based video codecs this data must begin and end on packet boundaries (no partial packets), and must contain enough video packets to decode exactly one image (either a keyframe or delta frame). Note: Foxglove does not support video streams that include B frames because they require lookahead.
format string Video format. Supported values: h264, h265, vp9, av1

FrameTransform

A transform between two reference frames in 3D space

Field Type Label Description
timestamp google.protobuf.Timestamp protobuf Timestamp (int seconds & nanos)
parent_frame_id string
child_frame_id string
translation Vector3 meters
rotation Quaternion

FrameTransforms

An array of transforms, each of which can transform points from a child frame to its parent frame

Field Type Label Description
transforms FrameTransform repeated

Quaternion

A quaternion representing a rotation in 3D space

Field Type Label Description
x double
y double
z double
w double

Vector3

A vector in 3D space that represents a direction only

Field Type Label Description
x double
y double
z double

Downloads last month
21