homer-v2

# HomER v2: Home Egocentric Robotics Dataset HomER v2 is a curated dataset of egocentric household activity videos designed for robotics, embodied AI, and video understanding research. The dataset contains 765 first-person videos representing approximately 100 hours of real-world household activities collected from 420 unique participants across 49 countries spanning 5 continents. Each video is accompanied by structured metadata and a natural-language description of the performed task. ## Dataset Contents Each record contains: - Unique video identifier - Activity category - Natural language task description - Video preview URL - Video download URL - Video duration - Additional metadata ## Dataset Composition The dataset contains 765 first-person videos representing approximately 100 hours of real-world household activities distributed across 10 activity categories. | Activity Category | Videos | Hours | | --- | ---: | ---: | | Indoor Plant Activities | 85 | 9.49 | | Cleaning | 95 | 9.98 | | Dining Table Tasks | 88 | 10.01 | | Personal Care & Bathroom Activities | 86 | 10.03 | | Desk & Office Tasks | 78 | 10.01 | | Laundry & Garment Care | 75 | 9.94 | | Home Organization | 69 | 9.99 | | Repair & Assembly | 67 | 10.09 | | Food Preparation | 66 | 10.06 | | Crafts & Fine Manipulation | 56 | 9.99 | | **Total** | **765** | **99.60** | ## Dataset Characteristics - Egocentric (first-person) recordings - Approximately 100 hours of video data - Data collected from 420 unique participants across 49 countries - Geographic coverage spanning Africa, Asia, Europe, North America, and South America - Natural language task descriptions for every video - Real-world household environments and activities - Balanced distribution across activity categories - Diverse object manipulation and interaction scenarios - Suitable for robotics, embodied AI, video understanding, and multimodal model evaluation ## Intended Use HomER v2 is intended for: - Robotics perception research - Embodied AI evaluation - Video understanding benchmarks - Activity recognition - Human-object interaction analysis - Video-language model evaluation - Task understanding and reasoning ## Data Format The dataset is distributed as structured metadata with references to source videos. Video assets are accessible through the provided preview and download URLs. ## Curated by **Bakyt Yrysov** (Toloka) **Oleksii Moga** (Toloka) ## Attribution This dataset is released under the **CC BY 4.0** license. If you use HomER v2, please provide attribution to Toloka and cite the dataset in publications, benchmarks, and derivative works. ## Citation ```bibtex @dataset{homer_v2_2026, title = {HomER v2: Home Egocentric Robotics Dataset}, author = {Toloka}, year = {2026}, publisher = {Hugging Face} } ```