Skip to content

Datasets for training a two-staged MLP network to estimate ground reaction force in a two-DoF point-foot robot.

Notifications You must be signed in to change notification settings

JaeHoon-An/Datasets-GRF-Estimation

Repository files navigation

Datasets-GRF-Estimation

Datasets for training a two-staged MLP network to estimate ground reaction force in a two-DoF point-foot robot.

Cite this repository: DOI

A. Datasets for GRF estimation (First stage MLP).csv

In the above CSV file, each collum denotes the following states of the robot.

1: Current Position of Hip Pitch Joint

2: First histroy of Position of Hip Pitch Joint

3: Second histroy of Position of Hip Pitch Joint

4: Current Position of Knee Pitch Joint

5: First histroy of Position of Knee Pitch Joint

6: Second histroy of Position of Knee Pitch Joint

7: Current Velocity of Hip Pitch Joint

8: First histroy of Velocity of Hip Pitch Joint

9: Second histroy of Velocity of Hip Pitch Joint

10: Current Velocity of Knee Pitch Joint

11: First histroy of Velocity of Knee Pitch Joint

12: Second histroy of Velocity of Knee Pitch Joint

13: Current Torque of Hip Pitch Joint

14: Current Torque of Knee Pitch Joint

15: Current Ground Reaction Force (Ground-truth)

B. Datasets for Sim2Real transfer (Second stage MLP).csv

In the above CSV file, each collum denotes the following states of the robot.

1: Estimated Ground Reaction Force from the First Stage MLP Network

2: Measured Ground Reaction Force (Ground-truth)

About

Datasets for training a two-staged MLP network to estimate ground reaction force in a two-DoF point-foot robot.

Resources

Stars

Watchers

Forks

Packages

No packages published