Ropedia raises $22M to scale HOMIE rigs for robot training data

The interesting line in Ropedia's funding announcement is not the $22 million, it is the founder's framing of the problem. According to SiliconANGLE's report, CEO Zhaoxi Chen said flatly that "there's not yet going to be a massive deployment of robots. We need to unlock the ChatGPT moment for robotics first." The Singapore company's whole pitch is that the missing ingredient for that moment is data, and that the cheapest way to get it is to strap a small camera rig onto ordinary people going about their day.

The rig is called HOMIE, short for Human-centric Omni Interaction and Experience. It is a head-mounted wearable with four directional cameras that weighs less than half a pound, and the new capital is meant to push production toward 10,000 units, expand US operations, and hire across hardware, software, AI infrastructure, and business development. Chen described the ambition in industrial terms, saying infrastructure means "you can push the boundary of data production from, let's say, 1,000 hours [to] 1 million hours of data production."

Why this matters if you are not building humanoids yourself: the robotics field has been arguing for a couple of years about whether the next unlock comes from simulation, teleoperation, or real human video, and each path implies a different supply chain. Ropedia is planting a flag on the human-video path with a proof point, its claim that it supplied more than half the training data behind the Allen Institute for AI's MolmoMotion 3D motion-forecasting model. If that pattern repeats with other foundation model teams, the data vendor layer starts to look a lot like what Scale AI became for text and images.

The honest caveat is that the reporting does not name a single investor in the round, does not disclose what a HOMIE unit costs to build or how Ropedia prices the resulting datasets, and points to only one named model customer. It is also an open technical question how well footage captured from a human head transfers to a robot with a very different body. What the announcement does give you is a specific bet, from a specific founder, that the constraint on embodied AI is boring and industrial, and that whoever wires up the capture pipeline first gets to sell picks and shovels to everyone else.