Embodal
Embodal provides first-person video datasets of professionals performing real paid work—cleaning, repair, and assembly—in customers' homes across Europe, the Gulf, and North America. The company captures whole jobs including setup, retries, failures, and cleanup, with four layers of consent per recording. Data is delivered in RLDS, WebDataset, or HDF5 formats through a five-step project workflow.
- Artificial Intelligence
- Robotics
Funding
Founders
Founder details are not available yet.
Product
Problem
Training embodied AI systems for household tasks requires video demonstrations of real work, but most available footage is either staged for cameras, lacks full task coverage, or fails to capture the messy, non-linear reality of professional jobs. This limits the effectiveness and safety of AI models trained on such data, particularly in domains like cleaning, repair, and assembly.
Solution
First Person Video provides first-person video datasets of professionals performing real paid work—cleaning, repair, assembly—in customers' homes across Europe, the Gulf, and North America. The company captures whole jobs, including setup, retries, failures, and cleanup, rather than polished takes staged for cameras. Every recording carries four layers of consent captured at marketplace, worker, household, and episode levels. Projects run through a five-step process—scope, pilot, review, scale, and deliver—ensuring data meets client acceptance criteria before full-scale collection. Final datasets are delivered in RLDS, WebDataset, or HDF5 formats, ready for integration into training pipelines with complete consent records.
Target Audience
Primary customers are AI research teams and robotics companies developing embodied models for household tasks, particularly those training on real-world first-person video in cleaning, repair, and assembly domains.
Features
- Whole-job capture including setup, retries, failures, and cleanup—not edited highlights
- Four-layer consent architecture: marketplace, worker, household, and episode level
- Five-step project workflow: scope, pilot, review, scale, and deliver with agreed acceptance criteria
- Data delivered in RLDS, WebDataset, or HDF5 formats with consent records for every recording
- Recurring batches from the same network of professionals and homes for longitudinal consistency
- Coverage across Europe, the Gulf, and North America with diverse household environments