umi
Datasets
All datasets matching “umi”HiFi-UMI-2K
HiFi-UMI-2K: High-Fidelity Robot-Free Manipulation Data
2,000 hours released · 6 synchronized camera views · 480+ scenes · 3 mm pose accuracy · <40 µs synchronization
🌐 Project Website |
📦 Dataset |
📄 Paper: arXiv:2607.25895
Examples from the HiFi-UMI corpus. Click the image to play the video.
📚 Introduction
HiFi-UMI is a portable, high-fidelity bimanual capture system for collecting robot-free manipulation demonstrations.… See the full description on the dataset page: https://huggingface.co/datasets/simple-world-lab/HiFi-UMI-2K.ROPE
Dataset Card for ROPE
The dataset used in this study is designed to evaluate and analyze multi-object hallucination by leveraging existing panoptic segmentation datasets. Specifically, it includes data from MSCOCO-Panoptic and ADE20K, ensuring access to diverse objects and their instance-level semantic annotations.
For more information, please visit Multi-Object Hallucination.
Dataset Construction
The dataset is divided into several subsets based on the distribution… See the full description on the dataset page: https://huggingface.co/datasets/sled-umich/ROPE.ICL_UMI_16jan_pickleUMI-Benchmark-v1
UMI-Benchmark-v1
UMI-Benchmark-v1 contains 20,000 real-world robot manipulation sessions across 10 tasks:
4 single-arm tasks and 6 dual-arm tasks.
Dataset Summary
Category
Tasks
Sessions
Single-arm
4
9,988
Dual-arm
6
10,012
Total
10
20,000
Tasks
ID
Folder
Type
Task
Sessions
T1
single_arm_task1
Single-arm
Stack Baskets
3,000
T2
single_arm_task2
Single-arm
Trash Bag
1,991
T3
single_arm_task3
Single-arm
Stamp Ink
2… See the full description on the dataset page: https://huggingface.co/datasets/UMIbenchmark/UMI-Benchmark-v1.MindCraftumi-robots
umi-robots
Part of umi, an open web crawl published as Parquet. Before you use any of this, read the exclusion list at open-index/umi-meta and filter the rows it names. Published files are never rewritten, so the exclusion list is how a takedown reaches you, and applying it is a condition of using the data rather than a suggestion.
One row per robots.txt fetch: the host, when we asked, what the origin answered, the raw text if it served one, and the summary our parser read out… See the full description on the dataset page: https://huggingface.co/datasets/open-index/umi-robots.