ABCDEFGHIJKLMNOPQRSTUVWXYZ
1
Phases:
2
intialStarting pose of the robot arm at the very beginning of a grasp episode.
3
approach
The first move in the grasp episode. Begins at the initial pose, and ends with the gripper in the workspace. This phase is separated from the other grasp phases, since the gripper is not visible at the beginning. During data collection, the target gripper pose for the approach phase was typically selected randomly.
4
grasp
Main grasp phase consisting of multiple steps. Each step corresponds to recording an image, making a decision about the next pose, and beginning to execute a move to that pose. At the end of the last step, the gripper is closed.
5
grasp/NThese sequence identifiers refer to steps in the grasp phase.
6
gripperClose the gripper at the end of the grasp episode.
7
withdraw
Withdraw the closed gripper from the bin vertically upward, roughly 10 cm above the bin surface.
8
move_to_present
After the withdraw phase, the robot moves the gripper in front of the camera to get a clean, unobstructed view of the object (this phase was skipped for some of the episodes and may not always be present).
9
presentPresentation of grasped object to camera.
10
raise
After presentation (or after withdraw, if present is skipped), the object is moved to a random position about 10 cm above the bin to drop the object.
11
dropOpen the gripper and drop the object (if any object is held) into the bin.
12
13
14
Feature Top Level Names:Phase
The top level feature names describe an entity (e.g. robot or camera) or a phase or step of the action (e.g. grasp, approach).
15
approachapproach
16
approachapproachPose reached at the end of the approach phase.
17
approach_sequenceapproach
Recorded robot status. Each sequence corresponds to robot telemetry (end effector poses, joint angles, recorded and commanded torques) recorded at about 20 Hz for the duration of that phase.
18
camera
4x4 transformation matrix from the base of the robot to the camera. This can be used if a camera calibration is needed. Note that camera calibration was not used in the closed-loop controller described in the paper, but is available for baselines.
19
graspgraspCommanded (target) pose for the last step in the grasp episode.
20
graspgrasp
21
graspgraspPose reached at the end of a step in the grasp episode.
22
grasp_sequencegrasp
Recorded robot status. Each sequence corresponds to robot telemetry (end effector poses, joint angles, recorded and commanded torques) recorded at about 20 Hz for the duration of that phase.
23
grippergripper
24
gripper_sequencegripper
Recorded robot status. Each sequence corresponds to robot telemetry (end effector poses, joint angles, recorded and commanded torques) recorded at about 20 Hz for the duration of that phase.
25
initialinitial
26
move_to_present_sequencemove_to_present
Recorded robot status. Each sequence corresponds to robot telemetry (end effector poses, joint angles, recorded and commanded torques) recorded at about 20 Hz for the duration of that phase.
27
num_grasp_stepsgrasp
Total number of steps (decisions) in the grasp episode, i.e. grasp/N has N in [0,...,num_grasp_steps - 1]
28
post_dropdropImages recorded after withdraw and raise but before the drop.
29
post_graspgraspImages recorded right after the gripper closes.
30
pregraspgraspImages recorded right before the gripper closes.
31
presentpresentImages recorded during the presentation to the camera.
32
present_sequencepresent
Recorded robot status. Each sequence corresponds to robot telemetry (end effector poses, joint angles, recorded and commanded torques) recorded at about 20 Hz for the duration of that phase.
33
raise_sequenceraise
Recorded robot status. Each sequence corresponds to robot telemetry (end effector poses, joint angles, recorded and commanded torques) recorded at about 20 Hz for the duration of that phase.
34
graspgrasp
Reached (actual) pose for the last step in the grasp episode. May differ from the commanded pose due to imperfect end-effector control at high speed.
35
robot
Identifying information about the robot used to collect the data. Typically a robot ID number, which can be used to figure out which datapoints were collected by the same robots.
36
withdraw_sequencewithdraw
Recorded robot status. Each sequence corresponds to robot telemetry (end effector poses, joint angles, recorded and commanded torques) recorded at about 20 Hz for the duration of that phase.
37
38
Feature Types:The feature type describes the organization and interpretation of data within the feature.
39
transforms/base_T_endeffector/vec_quat_7
A pose is a 6 degree of freedom rigid transform represented with 7 values: vector (x, y, z) and quaternion (x, y, z, w). A pose is always annotated with the target and source frames of reference. For example, base_T_camera is a transform that takes a point in the camera frame of reference and transforms it to the base frame of reference.
40
commanded_pose
Commanded pose is the input to the inverse kinematics computation, which is used to determine desired joint positions.
41
reached_posePose calculated by forward kinetics from current robot state in status updates.
42
joint/commanded_torques
Commanded torques are the torque values calculated from the inverse kinematics of the commanded pose and the robot driver.
43
joint/external_torques
External torques are torques at each joint that are a result of gravity and other external forces. These values are reported by the robot status query.
44
joint/positionsRobot joint positions as reported by the robot status query.
45
joint/velocitiesRobot joint velocities as reported by the robot status query.
46
depth_image
Depth image is encoded as an RGB PNG image where the RGB 24-bit value is an integer depth with scale 1/256 of a millimeter.
47
imageCamera RGB images are stored in JPEG format.
48
params
This is a simplified representation of a commanded robot pose and gripper status. These are the values that were solved for in the network.
49
50
51
Feature Keys
52
approach/params
53
approach/transforms/base_T_endeffector/vec_quat_7
54
approach/reached_pose/transforms/base_T_endeffector/vec_quat_7
55
approach_sequence/joint/commanded_torques
56
approach_sequence/transforms/base_T_endeffector/vec_quat_7
57
approach_sequence/joint/external_torques
58
approach_sequence/joint/positions
59
approach_sequence/joint/velocities
60
camera/transforms/camera_T_base/matrix44
61
grasp/commanded_pose/transforms/base_T_endeffector/vec_quat_7
62
grasp/0/depth_image/encoded
63
grasp/1/depth_image/encoded
64
grasp/2/depth_image/encoded
65
grasp/3/depth_image/encoded
66
grasp/4/depth_image/encoded
67
grasp/5/depth_image/encoded
68
grasp/6/depth_image/encoded
69
grasp/7/depth_image/encoded
70
grasp/8/depth_image/encoded
71
grasp/9/depth_image/encoded
72
grasp/10/depth_image/encoded
73
drop/commanded_pose/transforms/base_T_endeffector/vec_quat_7
74
grasp/depth_image/encoded
75
grasp/0/params
76
grasp/1/params
77
grasp/2/params
78
grasp/3/params
79
grasp/4/params
80
grasp/5/params
81
grasp/6/params
82
grasp/7/params
83
grasp/8/params
84
grasp/9/params
85
grasp/10/params
86
grasp/0/commanded_pose/transforms/base_T_endeffector/vec_quat_7
87
grasp/1/commanded_pose/transforms/base_T_endeffector/vec_quat_7
88
grasp/2/commanded_pose/transforms/base_T_endeffector/vec_quat_7
89
grasp/3/commanded_pose/transforms/base_T_endeffector/vec_quat_7
90
grasp/4/commanded_pose/transforms/base_T_endeffector/vec_quat_7
91
grasp/5/commanded_pose/transforms/base_T_endeffector/vec_quat_7
92
grasp/6/commanded_pose/transforms/base_T_endeffector/vec_quat_7
93
grasp/7/commanded_pose/transforms/base_T_endeffector/vec_quat_7
94
grasp/8/commanded_pose/transforms/base_T_endeffector/vec_quat_7
95
grasp/9/commanded_pose/transforms/base_T_endeffector/vec_quat_7
96
grasp/10/commanded_pose/transforms/base_T_endeffector/vec_quat_7
97
grasp/0/reached_pose/transforms/base_T_endeffector/vec_quat_7
98
grasp/1/reached_pose/transforms/base_T_endeffector/vec_quat_7
99
grasp/2/reached_pose/transforms/base_T_endeffector/vec_quat_7
100
grasp/3/reached_pose/transforms/base_T_endeffector/vec_quat_7