Robots are getting higher at seeing, talking, and planning. What they nonetheless battle with is the messy enterprise of doing.
For a machine to reliably pack a field, wipe a desk, load a warehouse shelf or help somebody at residence, it wants greater than web textual content and video clips. It wants to grasp grip, weight, timing, motion, and context — the type of bodily judgement people construct by means of repeated expertise. Singapore-based Ropedia is betting that this lacking layer will grow to be one of the crucial necessary infrastructure markets in synthetic intelligence (AI).
Additionally Learn: “Information, not {hardware}, is the actual bottleneck in humanoids”: Matrix Robotics CEO Allen Zhang
The startup has raised US$22 million in pre-Collection A funding, taking its complete funding to US$30 million. The spherical was backed by enterprise traders targeted on AI, deeptech and infrastructure in Southeast Asia, although the corporate didn’t disclose particular investor names.
A earlier spherical included traders and tremendous angels related to Google, Andreessen Horowitz, NVIDIA, and Amazon.
Ropedia will use the capital to increase its real-world information assortment operations into Southeast Asia and North America, develop its Singapore and US groups, and enhance manufacturing of its wearable seize {hardware}. It additionally plans to strengthen its information platform with annotation instruments, high quality analytics, and compliance programs, whereas increasing analysis into information basis fashions and world fashions, AI programs designed to construct an inner understanding of how the bodily world behaves.
Why bodily AI wants completely different information
The surge of curiosity in “bodily AI” follows the speedy advances made by giant language fashions. The essential concept is to convey AI out of screens and into machines that may function in the actual world, from industrial robots and autonomous autos to humanoids and residential assistants.
However robots face a more durable information downside than chatbots. Textual content-based AI programs had been skilled on monumental volumes of written materials already obtainable on-line. Robotics information is scarcer, costlier to gather, and way more depending on context. A video of an individual lifting an object might present the motion, however not all the time the drive used, the hand movement, the depth of the scene or the refined changes made alongside the way in which.
That’s the place Ropedia is positioning itself. Its platform captures what the corporate calls multimodal human expertise information: selfish video, depth, movement, and audio collected by means of proprietary wearable {hardware}. The info is then synchronised, processed and transformed into datasets that robotics and embodied AI builders can use to coach their fashions.
Additionally Learn: Rise of the machines: 20 robotics startups shaping Southeast Asia’s future
“A robotic can’t play baseball by watching a video any greater than you may be taught to journey a motorcycle by studying about it,” mentioned Zhaoxi Chen, CEO and co-Founding father of Ropedia. “The robotic should perceive what it’s wish to grip a bat and know the timing it takes to hit a ball.”
That rationalization will get to the guts of the problem. The subsequent stage of robotics isn’t just about recognition; it’s about interplay. Machines want coaching information that information how people transfer by means of kitchens, workshops, factories, places of work and streets, and the way these actions change throughout cultures, layouts and environments.
A Singapore base for a world robotics information play
Based within the second half of 2025, Ropedia is headquartered in Singapore and likewise has an workplace in Mountain View, California. Its founding group combines educational analysis and trade expertise in laptop imaginative and prescient and embodied AI.
Chen’s work spans 3D laptop imaginative and prescient, generative basis fashions and multimodal content material technology. CTO Fangzhou Hong beforehand labored on Meta’s selfish multimodal intelligence analysis, whereas Chief Scientist Ziwei Liu is an Affiliate Professor at Nanyang Technological College in Singapore.
That Singapore connection issues. Southeast Asia is turning into a helpful testbed for bodily AI due to its mixture of superior manufacturing, logistics hubs, dense city environments and service-heavy economies. Singapore, specifically, has pushed robotics in sectors similar to healthcare, cleansing, logistics and meals companies, partly due to labour constraints and its high-cost working surroundings.
The area additionally gives environmental variety that robotics corporations can’t simply replicate in a lab — humid warehouses, crowded retail areas, blended transport programs and assorted family settings.
Additionally Learn: dConstruct lands US$125M Collection A to scale robotics for GPS-denied environments
For Ropedia, increasing information assortment in Southeast Asia may assist its prospects prepare fashions which might be much less brittle when deployed outdoors managed environments. A robotic skilled solely on neatly staged manufacturing unit or residence information from one geography might fail when confronted with completely different lighting, room layouts, instruments, packaging, languages or consumer behaviour.
The corporate claims its method can scale back data-collection prices by as much as 50x in contrast with conventional strategies. Its wearable gadget, known as HOMIE, has entered mass manufacturing to help bigger deployments. Ropedia says it already serves greater than 20 robotics and basis mannequin corporations throughout North America, China and Singapore, in areas together with embodied AI and spatial intelligence.
Its flagship dataset, Xperience-10M, is described by the corporate as one of many world’s largest human expertise datasets. Ropedia additionally gives customized Information-as-a-Service merchandise for robotics and embodied AI builders that want task-specific or geography-specific information.
The aggressive discipline
Ropedia is getting into a market that’s nonetheless forming, however not empty. Its rivals are prone to come from a number of instructions. Information-labelling and AI infrastructure corporations similar to Scale AI, Appen, and TELUS Worldwide AI have lengthy served machine studying groups, although a lot of their work has targeted on labelling quite than capturing bodily interplay information at supply.
Artificial information corporations similar to Datagen have focused laptop imaginative and prescient and simulation use circumstances, providing one other solution to prepare fashions when real-world information is scarce. In the meantime, robotics and embodied AI startups similar to Bodily Intelligence, Skild AI, and Determine AI are constructing their very own mannequin and information pipelines, which may scale back their reliance on outdoors suppliers.
Ropedia’s guess is that unbiased, large-scale, real-world human expertise information will grow to be a shared infrastructure layer, very similar to cloud infrastructure did for software program startups.
The query is whether or not robotics corporations will purchase that layer externally or proceed constructing it in-house. In AI, the reply has typically been blended: corporations outsource some infrastructure when it saves time, however preserve strategically delicate information shut. Ropedia should show not solely that its datasets are cheaper and broader, however that they’re dependable, compliant and meaningfully enhance mannequin efficiency.
That compliance layer might grow to be extra necessary as bodily AI leaves analysis labs. Information captured in houses, factories, and public areas can elevate privateness and consent points, particularly when audio, video, and motion information are concerned. Southeast Asia’s regulatory surroundings is fragmented, with completely different information safety guidelines throughout markets, so any firm constructing regional information infrastructure will want cautious governance from the beginning.
Additionally Learn: Past productiveness: How AI could make work extra human
Nonetheless, the timing is beneficial. Traders are looking for the following infrastructure layer after the increase in giant language fashions, whereas robotics corporations are below stress to point out that their programs can transfer past demos. If Ropedia can flip human bodily expertise into structured, reusable coaching information, it may sit near the picks-and-shovels layer of the robotics economic system.
Chen frames the chance in infrastructure phrases: cloud computing wanted information centres, language AI wanted web textual content, and bodily intelligence will want real-world interplay information. The declare is bold, however the route of journey is obvious. If AI is to maneuver from answering inquiries to dealing with objects, opening doorways and dealing beside individuals, it might want to be taught from the bodily world, not simply take a look at it.
The put up Ropedia raises US$22M to construct the information layer for robots that perceive the actual world appeared first on e27.
