0418 GMT - Data collection is emerging as a critical factor for humanoid robotics in 2026, Nomura analysts say in a research note. For shipments of every 100,000 units of humanoid robots, that requires 10 million hours of data, they estimate. There are four different types of data collected to train robots including no-embodiment data, or human movements captured through wearable devices. Another type of data is collected via a human operator remotely controlling a physical robot while the system records synchronized data like camera feeds and joint states to train foundation models, they say. Nomura thinks a software-hardware closed loop--spanning collection, transmission, evaluation, training, deployment and debugging--is a structurally defensible business model for pure-play data vendors. (sherry.qin@wsj.com)
(END) Dow Jones Newswires
July 06, 2026 00:18 ET (04:18 GMT)
Copyright (c) 2026 Dow Jones & Company, Inc.