Dataset Viewer
Auto-converted to Parquet
Search is not available for this dataset
video
video
label
class label
2 classes
0tactile_video_gelsight
0tactile_video_gelsight
0tactile_video_gelsight
0tactile_video_gelsight
0tactile_video_gelsight
0tactile_video_gelsight
0tactile_video_gelsight
0tactile_video_gelsight
0tactile_video_gelsight
0tactile_video_gelsight
0tactile_video_gelsight
0tactile_video_gelsight
0tactile_video_gelsight
0tactile_video_gelsight
0tactile_video_gelsight
0tactile_video_gelsight
0tactile_video_gelsight
0tactile_video_gelsight
0tactile_video_gelsight
0tactile_video_gelsight
1tactile_video_mctac_v1
1tactile_video_mctac_v1
1tactile_video_mctac_v1
1tactile_video_mctac_v1
1tactile_video_mctac_v1
1tactile_video_mctac_v1
1tactile_video_mctac_v1
1tactile_video_mctac_v1
1tactile_video_mctac_v1
1tactile_video_mctac_v1
1tactile_video_mctac_v1
1tactile_video_mctac_v1
1tactile_video_mctac_v1
1tactile_video_mctac_v1
1tactile_video_mctac_v1
1tactile_video_mctac_v1
1tactile_video_mctac_v1
1tactile_video_mctac_v1
1tactile_video_mctac_v1
1tactile_video_mctac_v1

Dataset of Reactive Diffusion Policy

Contents

Description

This is the raw and postprocessed dataset used in the paper Reactive Diffusion Policy: Slow-Fast Visual-Tactile Policy Learning for Contact-Rich Manipulation.

Structure

We offer two versions of the dataset: one is the full dataset used to train the models in our paper, and the other is a mini dataset for easier examination. Both versions include raw and postprocessed subsets of peeling, wiping and lifting.

Each raw subset is structured as follows:

subset_name
β”œβ”€β”€ seq_01.pkl
β”œβ”€β”€ seq_02.pkl
β”œβ”€β”€ ...

Note that we split the full raw lifting subset into 2 parts due to file size restrictions.

Each postprocessed subset is stored in Zarr format, which is structured as follows:

 β”œβ”€β”€ action (25710, 4) float32
 β”œβ”€β”€ external_img (25710, 240, 320, 3) uint8
 β”œβ”€β”€ left_gripper1_img (25710, 240, 320, 3) uint8
 β”œβ”€β”€ left_gripper1_initial_marker (25710, 63, 2) float32
 β”œβ”€β”€ left_gripper1_marker_offset (25710, 63, 2) float32
 β”œβ”€β”€ left_gripper1_marker_offset_emb (25710, 15) float32
 β”œβ”€β”€ left_gripper2_img (25710, 240, 320, 3) uint8
 β”œβ”€β”€ left_gripper2_initial_marker (25710, 25, 2) float32
 β”œβ”€β”€ left_gripper2_marker_offset (25710, 25, 2) float32
 β”œβ”€β”€ left_gripper2_marker_offset_emb (25710, 15) float32
 β”œβ”€β”€ left_robot_gripper_force (25710, 1) float32
 β”œβ”€β”€ left_robot_gripper_width (25710, 1) float32
 β”œβ”€β”€ left_robot_tcp_pose (25710, 9) float32
 β”œβ”€β”€ left_robot_tcp_vel (25710, 6) float32
 β”œβ”€β”€ left_robot_tcp_wrench (25710, 6) float32
 β”œβ”€β”€ left_wrist_img (25710, 240, 320, 3) uint8
 β”œβ”€β”€ right_robot_gripper_force (25710, 1) float32
 β”œβ”€β”€ right_robot_gripper_width (25710, 1) float32
 β”œβ”€β”€ right_robot_tcp_pose (25710, 9) float32
 β”œβ”€β”€ right_robot_tcp_vel (25710, 6) float32
 β”œβ”€β”€ right_robot_tcp_wrench (25710, 6) float32
 β”œβ”€β”€ target (25710, 4) float32
 └── timestamp (25710,) float32

Usage

Follow the README in our GitHub repo to postprocess the raw data and train the model.

Tactile Dataset

We also provide the raw videos of the tactile dataset used for generate the PCA embedding in our paper.

Downloads last month
501

Models trained or fine-tuned on WendiChen/reactive_diffusion_policy_dataset