About the dataset

[zhanying3sui]

Very nice job. Could you share your datset? Thanks so much

[ACRoboT-buaa]

Thank you for your interest in our work. We apologize for the delay in responding to you.

dataset/robot.npz is the dataset file we generated. The dataset can also be found here.

[JianQin-Order]

Hello. When I clicked on the link (https://acrobot-buaa.github.io/pathplan.html) to download the dataset, it showed this message:
"version https://git-lfs.github.com/spec/v1
oid sha256:2f3dd751fbf7e7d3de7dffdfb53f61055ff4b04328cea8397ac00b8463c0b7f3
size 1411984514"

[ACRoboT-buaa]

I'm sorry, but this file is too large and exceeds the limit of free GitHub. We are working on it and will provide a new link in two to three days.

[alsike22]

Hello. When I clicked on the link (https://acrobot-buaa.github.io/pathplan.html) to download the dataset, it showed this message: "version https://git-lfs.github.com/spec/v1 oid sha256:2f3dd751fbf7e7d3de7dffdfb53f61055ff4b04328cea8397ac00b8463c0b7f3 size 1411984514"

Very nice job. Could you share your datset? Thanks so much

Thank you for your attention.
To avoid the limitation of LFS, we provide the dataset in npz and hdf5 formats on Google Drive instead. Please go there to download. Thank you.
Here's the link:
https://drive.google.com/drive/folders/1iJC03A8MZWJliMj2k0SltDjwI0yZQht8?usp=sharing

[JianQin-Order]

Thank you very much for sharing the dataset. I was wondering if it would be possible to provide some additional clarification regarding the data items. For example, from the shared data, I noticed that xtrain contains real-valued entries with a shape of 98000×254×7, while ytrain has a shape of 98000×23. Could you kindly provide more details about the meaning of each dimension? In particular, I’m interested in understanding the interpretation of the 7-dimensional Pose values for the 6-DOF robotic arm. I couldn’t seem to find a clear explanation of these data elements in the provided paper or code.

[alsike22]

Thank you for your questions.

Regarding the 7-dimensional Pose values: the dataset is collected from a 7-DoF redundant robotic arm — specifically, the Franka Emika Panda. Therefore, the 7-dimensional pose entries correspond to the joint angles of the seven actuated joints of the robot, rather than a 6-DoF end-effector pose.

Here's a more complete explanation of the dataset structure and how x_train and y_train relate to each other:

x_train (shape: 98000 × 254 × 7)

• x_train contains 98,000 motion trajectories of the Franka Panda 7-DoF robotic arm.
• Each trajectory consists of 254 frames, representing a continuous motion sequence.
• Each frame is a 7-dimensional vector, where each value corresponds to the rotation angle of one of the robot’s 7 joints.

In other words, x_train[i] is the optimal motion trajectory for the i-th robot task.

y_train (shape: 98000 × 23)
y_train contains the task definitions corresponding to each trajectory in x_train.

Each row y_train[i] defines a specific robot task, and x_train[i] gives the corresponding optimal trajectory to complete that task.

The 23 elements of y_train[i] are interpreted as follows:

• y_train[i][0:7]: Initial joint angles (start pose).
• y_train[i][7:14]: Final joint angles (end pose).
• y_train[i][14:17]: XYZ coordinates of the first obstacle center.
• y_train[i][17:20]: XYZ coordinates of the second obstacle center.
• y_train[i][20:23]: XYZ coordinates of the final target point for the end-effector.

All scenes contain two spherical obstacles, and all positions are normalized within the robot’s workspace.

We’ll also update the README to clarify these details in the dataset description.
Please feel free to reach out if you have any other questions!