Atlas

6thSense builds tactile capture hardware and collects multimodal human demonstration data—combining touch, vision, depth, force, and hand pose—to help train dexterous robots.

Asimov collects first-person video of people doing everyday tasks in real environments and sells annotated datasets to robotics teams training humanoid robots.

Build AI says it is building "high quality egocentric datasets for robotics & physical AI," including what it calls the largest egocentric data collection effort in history…

Claru is a training-data platform for physical AI that collects and enriches egocentric video, robotics trajectories, world-model footage, and expert human judgments for robotics and embodied-AI…

Cortex AI builds large-scale real-world robot and egocentric human datasets for embodied AI, including workplace first-person video, robot trajectories, and human-in-the-loop evaluations.

EgoScale is a robotics data company that provides large-scale, real-world egocentric human demonstrations and structured training data for VLA models, world models, and embodied AI.

FPV Labs is a research and product company building infrastructure for embodied AI, including Stera, a platform for capturing, processing, evaluating, and exporting egocentric multimodal human…

Hub is a distributed real-world data infrastructure company that sources and delivers rights-cleared multimodal training data for frontier AI labs and robotics companies.

Human Archive is a data lab that collects, labels, and synchronizes multimodal human behavior data to train robotics and physical-AI models. (humanarchive.ai)

Instawork is a staffing marketplace that connects businesses with hourly workers, and it now also markets a robotics initiative that pays workers to record real-world task demonstrations for robot…

Luel is a rights-cleared multimodal training-data marketplace and collection engine that supplies custom and off-the-shelf datasets to frontier AI teams.

Maxinsights collects and structures recordings of people doing real world tasks to create training data for physical AI and robotics models. (maxinsights.ai)

Mecka AI builds the data and deployment layer for physical AI, collecting and structuring real-world human activity data so robotics and embodied-AI teams can train and deploy systems more reliably.

Neocambrian AI builds data-collection infrastructure for physical AI/robotics, capturing large-scale human action data to train embodied AI systems.

Nferent AI builds Physical AI data infrastructure by capturing, labeling, and delivering real-world human skill datasets for robotics and humanoid AI companies, using multi-modal collection across…

OpenGraph Labs builds hardware and software for high-fidelity robotics data, including multimodal capture systems and tactile gloves for contact-rich robot learning.

Praxis captures egocentric video, 3D scans, and other multimodal data from real industrial and residential environments to supply training data for physical AI and humanoid robotics.

Renlei Labs builds and sells egocentric human activity datasets for robotics and AI, capturing first-person video, motion, and sensor data from real workers performing real tasks.

Roborecs is a Sofia-based company building EU-consented first-person human demonstration datasets for humanoid robot training, using head-mounted and wrist cameras plus multimodal sensors to…

Ropedia captures and structures people’s real world activity into multimodal training data for robotics and physical AI.

Shift is a robotics training-data company that pays people to record everyday household and trade tasks, then turns those recordings into anonymized datasets for AI and robotics customers.

Unidata is an AI training data company that provides data collection, expert labeling, and ready-made datasets, including egocentric first-person video and sensor data for robotics, AR/VR, and…

Waffle Video coordinates first person video capture with sensor data to create training datasets for robotics and physical AI teams. (wafflevideo.ai)