metadata
language:
- en
license: mit
tags:
- robotics
- test-time-adaptation
- reinforcement-learning
- continuous-control
- flow-matching
dataset_info:
features:
- name: seed
dtype: int32
- name: step
dtype: int32
- name: joint_torques
sequence: float32
Dataset Card for quadruped_reversible_flow
Dataset Description
This dataset contains Proportional-Derivative (PD) corrective joint torques generated during Test-Time Adaptation (TTA) simulations for the Quadruped environment using the reversible_flow policy. The dataset is structured for use in downstream machine learning workflows (such as training secondary diffusion or flow-matching models).
Physical Environment
- Robot: Quadruped
- Degrees of Freedom (DOF): 12
Normalization
The continuous joint_torques have been standardized globally to zero-mean and unit-variance across all seeds.
To un-normalize the data back to raw physical torque values, use the following constants:
- Mean ($\mu$):
[0.5545183420181274, -0.8843424916267395, 4.045175075531006, -0.5519294738769531, -0.8843178153038025, 4.045175552368164, 0.004045533481985331, 2.6992430686950684, 4.048221111297607, 0.08398226648569107, 2.698462724685669, 4.04833459854126] - Standard Deviation ($\sigma$):
[0.061799634248018265, 0.140830397605896, 0.167127788066864, 0.061556149274110794, 0.14076882600784302, 0.1671377569437027, 0.022350719198584557, 0.19783522188663483, 0.18377649784088135, 0.03169241175055504, 0.1978311985731125, 0.1838681399822235]
Usage
from datasets import load_dataset
dataset = load_dataset("elprofesoriqo/quadruped_reversible_flow")