Nomadic raises $8.4M to turn drone and robot video into training datasets

Startup Nomadic has raised $8.4 million in a seed round to tackle one of the autonomous transport industry's biggest infrastructure problems: what to do with the enormous volumes of video that self-driving cars and robots generate continuously. The company has built a deep learning platform that automatically converts raw camera footage from autonomous systems into structured, searchable datasets. Instead of spending months hand-labeling terabytes of recordings, developers of robots and self-driving vehicles get a ready-to-use annotated database.

The scale of the problem Nomadic is solving is easy to underestimate. A single autonomous car equipped with a standard sensor suite of cameras, lidars and radars can generate from one to several terabytes of data per hour. Multiply that by a fleet of hundreds or thousands of vehicles running in test and commercial programs worldwide, and the task comes into focus: we are talking about petabytes of footage that must be stored, processed and made usable for training the next generations of models.

The traditional approach, manual labeling through outsourcing platforms, works but is slow and expensive. Nomadic's new approach is to automate the whole process with its own deep learning model, trained to understand the context of scenes in automotive and robotic environments.

The practical result: engineers can do more than store video; they can work with it as a full-fledged database. You can enter a query such as "find all scenes where a pedestrian steps into the road in poor lighting" or "show cases where the car collided with an unrecognized obstacle" and get relevant clips in seconds rather than weeks of manual review.

An $8.4M round looks modest next to the billions invested in developing the autonomous systems themselves, but it reflects a growing understanding in the market: data infrastructure is as critical a component as the perception algorithms. Without effective management of training data, the development of self-driving technology hits an operational ceiling.

Nomadic works in a niche that is getting more competitive as the autonomous systems market matures. Interest in ML data management tools jumped after major players publicly acknowledged that training data quality is one of their main competitive advantages.

The category of data processing and labeling tools for autonomous systems is stepping out of the shadows. Where such companies once remained invisible suppliers to large players, investors are now starting to see standalone value in them, much as developer tools grew into a separate multibillion-dollar SaaS segment.

Nomadic is betting that the winner of this race will be whoever first gives self-driving developers an answer to the question: what do we do with all this video?