Robots are getting better at seeing, speaking, and planning. What they still struggle with is the messy business of doing.
For a machine to reliably pack a box, wipe a table, load a warehouse shelf or assist someone at home, it needs more than internet text and video clips. It needs to understand grip, weight, timing, movement, and context — the kind of physical judgement humans build through repeated experience. Singapore-based Ropedia is betting that this missing layer will become one of the most important infrastructure markets in artificial intelligence (AI).
Also Read: “Data, not hardware, is the real bottleneck in humanoids”: Matrix Robotics CEO Allen Zhang
The startup has raised US$22 million in pre-Series A funding, taking its total funding to US$30 million. The round was backed by venture investors focused on AI, deeptech and infrastructure in Southeast Asia, though the company did not disclose specific investor names.
A previous round included investors and super angels connected to Google, Andreessen Horowitz, NVIDIA, and Amazon.
Ropedia will use the capital to expand its real-world data collection operations into Southeast Asia and North America, grow its Singapore and US teams, and increase manufacturing of its wearable capture hardware. It also plans to strengthen its data platform with annotation tools, quality analytics, and compliance systems, while expanding research into data foundation models and world models, AI systems designed to build an internal understanding of how the physical world behaves.
Why physical AI needs different data
The surge of interest in “physical AI” follows the rapid advances made by large language models. The basic idea is to bring AI out of screens and into machines that can operate in the real world, from industrial robots and autonomous vehicles to humanoids and home assistants.
But robots face a harder data problem than chatbots. Text-based AI systems were trained on enormous volumes of written material already available online. Robotics data is scarcer, more expensive to collect, and far more dependent on context. A video of a person lifting an object may show the action, but not always the force used, the hand motion, the depth of the scene or the subtle adjustments made along the way.
That is where Ropedia is positioning itself. Its platform captures what the company calls multimodal human experience data: egocentric video, depth, motion, and audio collected through proprietary wearable hardware. The data is then synchronised, processed and converted into datasets that robotics and embodied AI developers can use to train their models.
Also Read: Rise of the machines: 20 robotics startups shaping Southeast Asia’s future
“A robot can’t play baseball by watching a video any more than you could learn to ride a bike by reading about it,” said Zhaoxi Chen, CEO and co-Founder of Ropedia. “The robot must understand what it’s like to grip a bat and know the timing it takes to hit a ball.”
That explanation gets to the heart of the challenge. The next stage of robotics is not just about recognition; it is about interaction. Machines need training data that records how humans move through kitchens, workshops, factories, offices and streets, and how those movements change across cultures, layouts and environments.
A Singapore base for a global robotics data play
Founded in the second half of 2025, Ropedia is headquartered in Singapore and also has an office in Mountain View, California. Its founding team combines academic research and industry experience in computer vision and embodied AI.
Chen’s work spans 3D computer vision, generative foundation models and multimodal content generation. CTO Fangzhou Hong previously worked on Meta’s egocentric multimodal intelligence research, while Chief Scientist Ziwei Liu is an Associate Professor at Nanyang Technological University in Singapore.
That Singapore connection matters. Southeast Asia is becoming a useful testbed for physical AI because of its mix of advanced manufacturing, logistics hubs, dense urban environments and service-heavy economies. Singapore, in particular, has pushed robotics in sectors such as healthcare, cleaning, logistics and food services, partly because of labour constraints and its high-cost operating environment.
The region also offers environmental diversity that robotics companies cannot easily replicate in a lab — humid warehouses, crowded retail spaces, mixed transport systems and varied household settings.
Also Read: dConstruct lands US$125M Series A to scale robotics for GPS-denied environments
For Ropedia, expanding data collection in Southeast Asia could help its customers train models that are less brittle when deployed outside controlled environments. A robot trained only on neatly staged factory or home data from one geography may fail when faced with different lighting, room layouts, tools, packaging, languages or user behaviour.
The company claims its approach can reduce data-collection costs by up to 50x compared with traditional methods. Its wearable device, called HOMIE, has entered mass production to support larger deployments. Ropedia says it already serves more than 20 robotics and foundation model companies across North America, China and Singapore, in areas including embodied AI and spatial intelligence.
Its flagship dataset, Xperience-10M, is described by the company as one of the world’s largest human experience datasets. Ropedia also offers custom Data-as-a-Service products for robotics and embodied AI developers that need task-specific or geography-specific data.
The competitive field
Ropedia is entering a market that is still forming, but not empty. Its competitors are likely to come from several directions. Data-labelling and AI infrastructure companies such as Scale AI, Appen, and TELUS International AI have long served machine learning teams, though much of their work has focused on labelling rather than capturing physical interaction data at source.
Synthetic data companies such as Datagen have targeted computer vision and simulation use cases, offering another way to train models when real-world data is scarce. Meanwhile, robotics and embodied AI startups such as Physical Intelligence, Skild AI, and Figure AI are building their own model and data pipelines, which could reduce their reliance on outside providers.
Ropedia’s bet is that independent, large-scale, real-world human experience data will become a shared infrastructure layer, much like cloud infrastructure did for software startups.
The question is whether robotics companies will buy that layer externally or continue building it in-house. In AI, the answer has often been mixed: companies outsource some infrastructure when it saves time, but keep strategically sensitive data close. Ropedia will have to prove not only that its datasets are cheaper and broader, but that they are reliable, compliant and meaningfully improve model performance.
That compliance layer may become more important as physical AI leaves research labs. Data captured in homes, factories, and public spaces can raise privacy and consent issues, especially when audio, video, and movement data are involved. Southeast Asia’s regulatory environment is fragmented, with different data protection rules across markets, so any company building regional data infrastructure will need careful governance from the start.
Also Read: Beyond productivity: How AI can make work more human
Still, the timing is favourable. Investors are hunting for the next infrastructure layer after the boom in large language models, while robotics companies are under pressure to show that their systems can move beyond demos. If Ropedia can turn human physical experience into structured, reusable training data, it could sit close to the picks-and-shovels layer of the robotics economy.
Chen frames the opportunity in infrastructure terms: cloud computing needed data centres, language AI needed internet text, and physical intelligence will need real-world interaction data. The claim is ambitious, but the direction of travel is clear. If AI is to move from answering questions to handling objects, opening doors and working beside people, it will need to learn from the physical world, not just look at it.
The post Ropedia raises US$22M to build the data layer for robots that understand the real world appeared first on e27.