Asimov's human-data bet: crowdsourced videos train humanoids; YC W26 standout

Asimov is a Y Combinator W26 startup building what its founders call the data infrastructure to enable the GPT-3 moment for humanoid robots. Founded by Anshul Verma, who worked at Scale AI and Amazon and did ML research at Berkeley, and Lyem Ningthou, who built data pipelines for the US Air Force, it pays people worldwide to record everyday tasks - cooking, cleaning, working - and packages the footage into datasets for robotics developers.

The bet is that humanoids are data-bound, not hardware-bound: language models trained on trillions of words and image models on billions of photos, but robots are starting from scratch. Existing robot datasets are narrow, largely factory footage repeating the same tasks in the same environments, while what robots need is diversity - thousands of people, in thousands of spaces, doing the full range of things humans do. Asimov's model is to capture that with phones people already own, backed by proprietary hardware kits and an annotation pipeline.

The startup launched publicly in mid-March 2026, just before W26 Demo Day, and TechCrunch named it one of the 16 most interesting companies of the ~190-strong cohort. By late March 2026, according to AC Studio, more than 5,000 contributors were participating and the company was supplying thousands of hours of human-motion data daily to major robotics labs.

In August 2026 Asimov's YC launch post said the largest robotics labs in the world were already customers, and that partners in restaurants and warehouses can cover workers' salaries by having them wear collection kits while they work - an explicit bet that paying ordinary people to record their own routines becomes the supply chain for humanoid training data.