Mecka CEO on why robotics needs its own data, not the internet

Josh Gao, co-founder and CEO of Mecka, discusses a fundamental challenge in robotics AI: large language models benefited from vast amounts of freely available internet data to pre-train on, but robotics has no equivalent. Robot training requires pixel-level visual data, action sequences, and logical reasoning tied to physical tasks—almost none of which exists in a form that foundation models can easily learn from.

Mecka's approach to solving this is vertical integration: the company builds the hardware wearables needed to collect training data, runs the data collection operations itself, and maintains strict quality control over every step. This end-to-end ownership of the data pipeline aims to solve the scarcity problem that has held back robot learning.