Singapore-based robotics data infrastructure firm Ropedia Pte. Ltd. today announced it raised $22 million in Pre-Series A funding to scale up its collection of real-world, multimodal interaction data to fuel artificial intelligence robotics models developed by technology companies.

Physical AI and embodied AI development is increasingly burdened by a lack of real-world data at a diversity and scale that robot-specific teleoperation struggles to produce. Developers in this field need models that can generalize across robots, objects and environments to scale products. Ropedia’s industry proposition is to gather human experience directly via video and convert it into model-ready multimodal datasets.

Although images and text exist in abundance, and there are many videos online, much of it cannot be readily curated or transformed into robotic trajectories that represent human-like activity. When a robot interacts with the real world, it needs synchronized, annotated, and curated information about movements, geometry and the consequences of those actions. Pictures of objects don’t work for this and unless video is well structured, it’s harder to concentrate into a clean dataset.

“There’s not yet going to be a massive deployment of robots,” founder and Chief Executive Zhaoxi Chen told SiliconANGLE in an interview. “We need to unlock the ChatGPT moment for robotics first.”

Chen described Ropedia’s approach to this challenge as existing in three layers: capture, annotation and deployment. The company focuses heavily on the annotation layer – curation, filtering, quality control and AI-assisted annotation – while more extensive fine-tuning and deployment loops build the long-term vision for the company.

From capture to clean data

To make this happen, the company built a wearable, head-mounted device called HOMIE, or Human-centric Omni Interaction and Experience. It’s shaped more or less like a crown that sits atop the head and mounts four cameras in the cardinal directions. It’s lightweight, at less than half a pound, and remains out of the user’s direct line of sight, making it easier to perform normal work or everyday activities without getting in their way.

The wearable captures raw experiences using video, while its processing backend synchronizes and annotates the resulting visual, spatial and motion data.

“This kind of ‘anywhere, by anybody’ is really valuable for enriching the diversity of the data,” Chen explained.

The company plans to release a second-generation version of the wearable that Chen characterized as both more attractive and lighter. It’s expected to be announced next month.

Ropedia looks to become similar to how cloud infrastructure became the foundation of internet services: largely invisible to end users, but necessary for ingesting, processing and serving information on an enormous scale. In terms of robotics and physical AI, that means becoming the go-to for multimodal synchronization, quality control, annotation and consistent data structures across as many contributors and environments as possible, delivered to a data-hungry industry.

“Infrastructure means that you can push the boundary of data production from, let’s say, 1,000 hours [to] 1 million hours of data production,” Chen said.

Despite the high visibility of HOMIE, Ropedia ultimately provides curated data produced by its pipeline. Customers can license existing datasets or use the company’s hardware to collect demonstrations within their own facilities. After that, Ropedia processes, curates and annotates the captured proprietary data.

Revolutionizing future robotics with curated data

The company’s data-centric foundation is already paying off. Chen noted that the Allen Institute for AI, a preeminent AI research company developing AI models, recently released MolmoMotion, a 3D motion forecasting model that can anticipate how objects will move within a scene. He explained that Ropedia supplied more than half of the data sources used to train the model.

Chen said the new funding will help Ropedia put its second-generation HOMIE hardware into large-scale production, with an eventual target of as many as 10,000 devices. The company also intends to hire across hardware, software, AI infrastructure and business development, including talent focused on efficiency and performance for its annotation models. Finally, the company intends to establish a larger United States branch and deepen relationships with North American customers and industry partners.

At its core, Ropedia provides clean data for robotics. This is an evolving industry supplying a growing number of robots moving from experiment onto factory floors. According to Chen, it’s an emerging discipline and it’s fundamental to the foundation that will allow embodied AI and smart robotics to scale up.

“Data is also a form of science,” Chen said. “It’s not only about collecting or accumulating the data. It’s more like how you organize and understand the flow between different domains.”

Image: SiliconANGLE/Microsoft Designer; Ropedia