Datasets:
ur3e_grasping_adipose_tissue
Private research dataset for UR3e imitation learning. Demonstrations were
recorded through the project's teleoperation pipeline for the task
grasp adipose tissue.
Dataset summary
- Robot:
ur3e - LeRobot format:
v3.0 - Episodes:
25 - Frames:
11025 - Frame rate:
20.0Hz - RGB input:
observation.images.endoscope - Depth input:
observation.images.endoscope_depth
The depth stream is stored as the LeRobot video feature observation.images.endoscope_depth and is decoded in metres for training.
Policy inputs
The intended policy input is RGB + depth + observation.state.
observation.state
| Index | Name |
|---|---|
| 0 | tcp_linear_x |
| 1 | tcp_linear_y |
| 2 | tcp_linear_z |
| 3 | forceps_open_ratio_raw |
| 4 | joint_6_roll |
| 5 | grasping_force_db |
action
| Index | Name |
|---|---|
| 0 | delta_position_x |
| 1 | delta_position_y |
| 2 | delta_position_z |
| 3 | delta_q6 |
| 4 | gripper_open_ratio |
The first four action values are the Cartesian delta and joint-6 roll command
used by the teleoperation DeltaCommand path. The fifth value is the gripper
open ratio and is represented separately by the deployment interface because
DeltaCommand has no gripper field.
Intended use and limitations
This dataset is intended for internal research and imitation-learning experiments on the UR3e setup. It is not a medical dataset, and it must not be used to make clinical decisions. Validate synchronization, image/depth quality, workspace limits, and robot safety before deploying a trained policy.
- Downloads last month
- 70