Egocentric Video Dataset — 100 Hours of Household Activities
This egocentric video dataset strengthens robotics workstreams with head-mounted ultrawide recordings of everyday household tasks, including meal prep, cleaning, organizing, and other daily routines. It spans short, atomic actions (about one minute) and longer, multi-step sequences (up to 20 minutes), preserving temporal context for modeling real-world behavior, object interaction, and task flow. Designed as an embodied AI and action recognition dataset, it supports training and evaluation where natural, in-home activity structure matters.
This egocentric video dataset strengthens robotics workstreams with head-mounted ultrawide recordings of everyday household tasks, including meal prep, cleaning, organizing, and other daily routines. It spans short, atomic actions (about one minute) and longer, multi-step sequences (up to 20 minutes), preserving temporal context for modeling real-world behavior, object interaction, and task flow. Designed as an embodied AI and action recognition dataset, it supports training and evaluation where natural, in-home activity structure matters.
This egocentric video dataset strengthens robotics workstreams with head-mounted ultrawide recordings of everyday household tasks, including meal prep, cleaning, organizing, and other daily routines. It spans short, atomic actions (about one minute) and longer, multi-step sequences (up to 20 minutes), preserving temporal context for modeling real-world behavior, object interaction, and task flow. Designed as an embodied AI and action recognition dataset, it supports training and evaluation where natural, in-home activity structure matters.
This egocentric video dataset strengthens robotics workstreams with head-mounted ultrawide recordings of everyday household tasks, including meal prep, cleaning, organizing, and other daily routines. It spans short, atomic actions (about one minute) and longer, multi-step sequences (up to 20 minutes), preserving temporal context for modeling real-world behavior, object interaction, and task flow. Designed as an embodied AI and action recognition dataset, it supports training and evaluation where natural, in-home activity structure matters.
Dataset specs
Type
Video
Content type
Robotics
Region/Locale
Various
Amount
100 hours
Leverage
Ideal for advancing robotics and embodied AI systems by learning real-world, first-person activity signals for task understanding and decision-making models, grounded in a robotics dataset.
Use cases
Train vision-based models to recognize and segment household activities from head-mounted footage, using a POV video dataset to improve action detection and task understanding in real-world environments.
Enhance robotic decision-making and navigation by leveraging egocentric sequences as AI training data to model human task execution, object interactions, and sequential behavior patterns.
Do you need a specific dataset?
We understand the uniqueness of every project. That's why we offer customizable dataset solutions to match your specific requirements.
Dataset specs
Type
Video
Content type
Robotics
Region/Locale
Various
Amount
100 hours