Ropedia raises $22 million for robotics training data pipeline

Ropedia Pte. Ltd. has raised $22 million in Pre-Series A funding to expand its robotics data infrastructure business, the Singapore-based company announced. The round gives the company more capital for a bet that embodied AI systems need better real-world training data before robot deployments can scale.

The company did not disclose its valuation, revenue, customer count or the investors in the round. Ropedia said the capital will support larger production of its second-generation HOMIE wearable device, hiring across hardware, software, AI infrastructure and business development, and expansion of its U.S. presence.

Ropedia is addressing a specific bottleneck in robotics: the shortage of synchronized, annotated data showing how people interact with objects and physical spaces. Text and image datasets are abundant, and video is widely available online, but most of that material is not structured in a way that can be converted into useful robot trajectories. Robotics models need data that ties motion, geometry and outcomes together in a form that can be used for training.

Founder and Chief Executive Zhaoxi Chen told SiliconANGLE that broad robot deployment is not imminent and that robotics first needs a breakout similar to what ChatGPT represented for language models. His view is that better data production, rather than only better models, is a prerequisite for that shift.

A wearable as data collection hardware

Ropedia’s main collection device is HOMIE, short for Human-centric Omni Interaction and Experience. The head-mounted wearable sits above the user’s line of sight and uses four cameras facing different directions to record everyday activities or work demonstrations. The device weighs less than half a pound, according to the company.

The hardware is only one part of the product. Ropedia’s backend synchronizes the recorded material and applies annotation across visual, spatial and motion data. Chen described the company’s system as having three layers: capture, annotation and deployment. He said Ropedia is putting particular emphasis on annotation, including curation, filtering, quality control and AI-assisted labeling.

The company plans to announce a second-generation HOMIE device next month. Chen said the new version is expected to be lighter and more visually polished. Ropedia’s longer-term production target is as many as 10,000 devices, though the company did not provide a timetable for reaching that number.

Data licensing and customer-owned collection

Ropedia’s commercial model is centered on curated datasets, not only hardware sales. Customers can license existing datasets from Ropedia or use the company’s devices to capture demonstrations in their own facilities. Ropedia then processes, filters and annotates that proprietary material.

The company is positioning itself as infrastructure for robotics data production, similar in role to how cloud services became a base layer for internet software. Chen told SiliconANGLE that the goal is to increase data production from thousands of hours to millions of hours. That is a useful ambition, but Ropedia has not disclosed current data volume, margins or how much of its pipeline is automated.

Ropedia has at least one visible research use case. Chen said the Allen Institute for AI used Ropedia-supplied material for more than half of the data sources behind MolmoMotion, a 3D motion forecasting model designed to predict how objects move within a scene.

The financing comes as embodied AI companies face the less glamorous side of robotics commercialization: model performance depends heavily on domain-specific data, and collecting that data is expensive and operationally messy. Ropedia is betting that standardized human-centric capture and annotation can become a sellable layer for robotics developers trying to move from lab demos into factories and other real environments.

This story draws on original reporting from SiliconANGLE.