There's a peculiar intimacy to the way Ropedia collects its data. Imagine: a crown-shaped rig perched on your head—four cameras staring outward, not inward—recording not just what you see, but how your hands move, where your eyes drift, the subtle choreography of your body navigating space. Less than half a pound, barely there. But always watching.
This is the hardware backbone of what Ropedia calls "physical AI infrastructure," and it's just netted the Singapore-based startup another $22 million. Announced July 23, 2026, the pre-Series A round pushes the company's total funding to $30 million—a notable haul for a venture incorporated just over a year ago, on July 11, 2025. An earlier $8 million tranche came through in March.
The company declined to name its backers, offering only that the round was led by "venture investors with deep experience in AI, enterprise tech and infrastructure" across Southeast Asia. Perhaps telling that even in disclosure, Ropedia keeps its cards close.
The Apparatus
At the heart of the operation sits HOMIE, the head-mounted device that feels less like typical consumer tech and more like something borrowed from a motion-capture studio. Those four outward-facing cameras work in concert, capturing synchronized streams: video, audio, depth mapping, hand and gaze tracking, body motion, camera pose. The synchronization tolerance? Sub-50 microseconds. That level of precision isn't about creating slick demos—it's about feeding machine learning models the kind of granular, multimodal data they'll need to interact with the physical world.
All that hardware outputs into Xperience-10M, a dataset Ropedia released publicly on Hugging Face. The numbers suggest scale: 10 million interaction episodes, 10,000 hours of first-person footage, roughly a petabyte of information. The Allen Institute for AI has already tapped Xperience videos for its MolmoMotion model, which debuted in June.
CEO Zhaoxi Chen, whose background spans 3D computer vision and multimodal AI, draws a clean historical parallel. "Text scraped from the internet was used to train the last generation of AI," he explained. "Real-world human experience, captured at the same scale, will train physical AI."
It's a tidy framing—though one that glosses over just how messy and expensive "real-world experience" is to capture, annotate, and maintain at internet scale.
Scaling the Factory
The fresh capital has a specific mandate: scale everything. Ropedia plans to expand data collection operations across Southeast Asia and North America, ramp HOMIE manufacturing to as many as 10,000 devices, build out annotation tooling and quality analytics infrastructure, and staff up in the United States. The company already operates offices in Mountain View, Kuala Lumpur, and Shanghai—a geographic spread that mirrors its customer base.
As of mid-year, Ropedia reportedly served more than 20 clients: robotics startups and foundation model companies scattered across North America, China, and Singapore. That's a respectable footprint for a company still in its first full year of operations, though the client list remains undisclosed.
The technical bench runs deep. Co-founder and CTO Fangzhou Hong earned his PhD at Nanyang Technological University and spent time at Meta Reality Labs Research. Chief Scientist Ziwei Liu holds an associate professorship at NTU, grounding the venture in Singapore's growing AI research ecosystem.
What Comes Next
Ropedia expects to ship a second-generation HOMIE device this August and move into large-scale production shortly after. In a recent interview, Chen hinted at the infrastructure ambitions driving the roadmap: pushing data production "from 1,000 hours to 1 million hours."
That kind of leap isn't just technical—it's operational, logistical, economic. Ropedia claims its approach slashes data collection costs by up to 50 times compared to traditional methods like teleoperation, where human operators manually control robots in real time. But the company hasn't disclosed pricing for either its hardware or dataset licensing, leaving questions about margins and accessibility.
For now, the bet remains straightforward: that the next frontier in AI—systems that move, manipulate, and navigate the physical world—will demand data infrastructure as robust as the text corpora that powered large language models. Whether Ropedia can build that infrastructure profitably, and at the scale Chen envisions, may determine if this $30 million gamble pays off.
