Asimov

Robotics teams lack enough real-world training data to teach robots how the physical world works. Most existing data comes from teleoperation, where a human remotely controls a robot in a lab, producing at most a few thousand hours of narrow, controlled footage. Meanwhile humans generate a vastly larger amount of real physical activity every day in kitchens, offices, hotels, and homes. Without diverse, messy, real-world data, robotics models cannot learn the variety of situations and interactions needed to generalize.

Asimov collects egocentric human activity data at scale using lightweight phone head mounts worn by individuals performing everyday tasks like cooking, cleaning, and running errands, as well as workers at partner businesses such as restaurants, hotels, and warehouses. The company handles the full pipeline: distributing hardware, training collectors, running quality assurance, and post-processing footage into datasets with 3D body pose, depth maps, semantic labels, and activity segmentation. Individuals are paid after their recordings pass QA, and partner businesses can have their workers' salaries covered in exchange for data, with revenue sharing available for larger partnerships. Asimov also operates its own cleaning company, Asimov Cleaning, where cleaners wear the recording kits while servicing apartments in the Bay Area, giving the company a direct channel for data collection.