Cervo Technologies

Robots create the most value where humans are hardest to train and most at risk, and those are exactly the places with no training data: mines, pipelines, oil and gas sites and data-centre builds. Cervo equips workers on shift with head, chest and wrist cameras, then has licensed electricians, miners and pipefitters annotate the footage so robot builders can train on it.

Around three million people die from their work every year, and the fatality rate in mining and oil and gas is several times the average. These are the jobs robots should take first, and they are the jobs with the least data to learn from, because nobody has been filming inside a trench, a substation or a mine shaft at scale, and because the footage that does exist needs an expert to label what is actually happening.

Cervo collects that data where it lives. Workers wear the rig on shift, the footage is de-identified, sliced and labelled with 3D hand tracking, and tradespeople who know the work correct a hazard-tuned model's labels. On top of the data the company builds benchmarks and reconstructed hazardous scenes that robots can train in without a site permit. The founders grew up around factories and power plants, and the company's first wedge is construction data from the environments where model builders want robots deployed first.