Karma-Ego is a unified, community-maintained index of egocentric and first-person video datasets for physical AI research. Discover what data exists, who maintains it, and how to access it — in one place.
Karma-Ego does not host or redistribute any dataset. We maintain structured metadata so researchers and engineers can find the right data and contact maintainers directly.
Filter datasets by task type, sensor modality, volume, license, or institution.
Each entry includes format specs, annotation details, paper citation, and access instructions.
Every listing links to the original project page and maintainer contact. No middleman, no friction.
Add a dataset by submitting a structured YAML entry to our GitHub repository.
Explore our curated collection of egocentric video datasets available for download and research.
Karma-Ego is open infrastructure. These organizations contribute datasets, research capacity, and real-world deployment context to the registry.
AI-powered staffing platform for blue-collar workers, Bengaluru. Contributing egocentric task and workflow datasets from real-world labor environments across India — a unique source of on-the-ground physical AI training data.
Research labs, robotics companies, and dataset maintainers welcome. Listing is always free.
If you maintain an egocentric or first-person video dataset, submit a pull request. Listing is free — gain visibility with researchers and robotics companies actively building physical AI systems.
Each dataset is a YAML file in
/registry/datasets/
name: Your Dataset Name institution: University / Research Lab / Company dataset_page: https://your-dataset-page.com volume_hours: 120 modalities: [rgb, depth, imu, gaze] license: CC BY 4.0 paper_url: https://arxiv.org/abs/xxxx.xxxxx access_url: https://your-dataset-page.com contact: maintainer@institution.edu country: Your Country task_type: [task name, activity type] data_type: egocentric video publisher: https://your-publisher-url.comSubmit on GitHub →
The physical AI research community lacks a single structured source of truth for egocentric video data. Karma-Ego is India's contribution to fixing that.
Robotics researchers spend weeks finding relevant datasets — scattered across lab pages, arXiv, and institutional portals with no common structure.
A community-maintained open registry. Structured metadata, direct maintainer contacts, and a searchable index — all in one GitHub repository.
Karma-Ego is fully open source. No sign-up, no API keys, no paywalls. Fork it, extend it, build on it.
Built in Bengaluru as India's contribution to the global physical AI and embodied intelligence research ecosystem.