https://doi.org/10.5061/dryad.tqjq2bw84

Description of the data and file structure

Please do not hesitate to contact Haonan Peng (penghaonan1993@gmail.com) if you need any help using this dataset.

All code about utilizing this dataset, including neural networks and regressions can be found at GitHub repo: 

https://github.com/HaonanPeng/Efficient-Data-driven-Joint-level-Calibration-of-Cable-driven-Surgical-Robots

as well as index, preprocessing.

Each recording .zip file contains training and testing trajectories as .csv files. "rand1200" in the file name indicates this is a random trajectory, otherwise indicating a zig-zag trajectory. "x", "y", or "z" in the file name indicates the direction of zig-zag trajectories, wherein "x - Joint 1", "y - Joint 2", "z - Joint 3", there can also be diagonal directions such as "xy - Joint 1-2". Numbers "05, 03, 025, 02, 016" in the file name means the sparsity of the zig-zag trajectory of 0.5, 0.333, 0.25, 0.2 and 0.1667. There may also be direction and/or sparsity notes for random trajectories. These notes for random trajectory indicate these random trajectories are recorded close to the zig-zag trajectories with these notes, which is to prevent the testing trajectories from being recorded too far away from the training zig-zag, causing a time effect.

• Recording 1 has training zig-zag trajectories with different direction, such as Joint 1, Joint 2, Joint 1-2, and so on.
• Recording 2 has training zig-zag trajectories with different sparsity.
• Recording 3 has has 6-hour testing trajectories in indleness, unloaded, and 500g loaded configuration, to study the time effects.
• Recording 4 has homing procedure during testing. And the 2 training sets, one has homing while one does not, to study the homing effects.

The column index of the dataset followed, more details can be found in our paper:

# left arm only ----------------------------

np.arange(1,2), # time stamp of ground truth 

np.arange(2,5), # ground truth joint position

np.arange(5,6), # time stamp of CRTK joint position, usually not used

np.arange(6,12), # joint position of CRTK joint position, usually not used, this is the same as the robot state joint position in the following sections

np.arange(12,13), # time stamp of robot state "ravenstate", all the following indice are from robot state

np.arange(13,21), # joint position

np.arange(29,30), # run level

np.arange(30,31), # sub-level

np.arange(31,32), # last sequence

np.arange(32,33), # arm type

np.arange(34,37), # end-effector position

np.arange(40,49), # end-effector orientation

np.arange(58,67), # end-effector orientation desired

np.arange(76,79), # end-effector position desired

np.arange(82,98), # encoder value

np.arange(98,106), # motor dac value

np.arange(114,122), # motor torque

np.arange(130,138) # motor position

np.arange(146,154), #  motor velocity

np.arange(162,170), # joint velocity

np.arange(178,186), # motor position desired

np.arange(194,202), # joint position desired

np.arange(210,211), # grasper desired

np.arange(212,228), # encoder offset

np.arange(228,234), # Jacobian velocity

np.arange(240,246) # Jacobian force

Code/software

Coda available at: https://github.com/HaonanPeng/Efficient-Data-driven-Joint-level-Calibration-of-Cable-driven-Surgical-Robots

Usage Note

This DRYAD dataset contains robot states and ground truth joint positions accompanying the 2024 manuscript "Efficient Data-driven Joint-level Calibration of Cable-driven Surgical Robots" by Peng, H et al., to be published in npj Robotics.

Access information

Data was derived from the following sources:

• RAVEN-II surgical robot