Datasetpublic

pravsels/arx5-mujoco-trajectories

ARX5 MuJoCo Trajectories Offline trajectory dataset for training world models on the ARX5 7-DOF robot arm in MuJoCo. Contents File Description train.pt PyTorch tensor file containing 1024 episodes of robot arm trajectories (~3.5M total frames) train_verification.json Dataset statistics: joint ranges, action deltas, coverage by stage and active joint count Dataset Stats Episodes: 1,024 Total frames: 3,547,018 Steps per episode: min… See the full description on the dataset page: https://huggingface.co/datasets/pravsels/arx5-mujoco-trajectories.

sourceHugging Facemitupdated 6mo agoView on Hugging Face

0likes16downloads

Dataset Card

ARX5 MuJoCo Trajectories

Offline trajectory dataset for training world models on the ARX5 7-DOF robot arm in MuJoCo.

File	Description
`train.pt`	PyTorch tensor file containing 1024 episodes of robot arm trajectories (~3.5M total frames)
`train_verification.json`	Dataset statistics: joint ranges, action deltas, coverage by stage and active joint count

Dataset Stats

—Episodes: 1,024
—Total frames: 3,547,018
—Steps per episode: min 1,027 / mean 3,464 / max 4,096
—State space: 14-dim (7 joint positions + 7 joint velocities)
—Action space: 7-dim (joint position targets)
—Discarded short episodes: 340
—Episodes with NaN: 0
—Stuck episodes: 0%
—Near joint-limit fraction: ~0.001%

Usage

python

import torch
data = torch.load("train.pt", weights_only=False)

Source

Generated from randomised MuJoCo rollouts of the ARX5 arm with multi-stage coverage (single joint, 2-3 joints, 4-7 joints active).