Ropedia’s Zhaoxi Chen on Building the Infrastructure Behind Physical AI

Artificial intelligence has largely been associated with language models and digital applications. The next wave is moving into warehouses, factories and hospitals, where machines have to work with real objects in real spaces. Building those machines takes a kind of data the internet does not hold.

Ropedia supplies that data. In this interview with International Business Weekly, Co-Founder and CEO Zhaoxi Chen explains why Physical AI is becoming an infrastructure market, where demand will come from first, and how the company plans to earn a long-term place in it.

1. Why Physical AI is the next business opportunity

“Physical AI brings intelligence into the activities that keep the economy running such as making products, moving goods, maintaining facilities, and helping people with everyday tasks,” Chen says.

Those industries already spend heavily on labor and equipment. AI that makes physical work more productive can therefore create value within budgets that already exist.

The opportunity grows as machines become easier to adapt to new tasks and environments. That could make automation viable for businesses that cannot justify engineering a separate system for every workflow.

“Our ambition is to build a globally leading Physical AI infrastructure company and support the many businesses developing and deploying intelligent machines.”

Zhaoxi Chen, Co-Founder and CEO of Ropedia

2. The infrastructure gap holding robots back

The data problem

For Chen, one of the biggest gaps is access to high-quality data that shows how physical work actually happens. A robot needs exposure to different objects, environments and ways of completing a task, and collecting that experience consistently is hard.

Ropedia’s answer

• HOMIE Gen2: a wearable capture device that records human activity from the wearer’s perspective, with synchronized visual and motion information
• The processing pipeline: adds structure, including spatial information and descriptions of actions, so developers can use the data
• Quality control and evaluation: the part of the stack that checks whether the data actually helps

“Companies need a dependable process for acquiring useful experience and checking whether it improves their systems,” Chen says.

3. Why robots need a different kind of data

A robot has to connect what it perceives with an action and that action’s physical consequences. Chen uses the example of picking up a cup: the robot must understand the cup’s position, choose a grasp, move without collision and respond if something slips. A useful training example has to preserve all of those relationships.

Internet text and video hold valuable knowledge, but they lack the precise spatial and action information that physical interaction needs. Robotics also needs experience of failure, recovery and variation across environments.

“This is why we emphasize information density alongside scale.”

Zhaoxi Chen, Co-Founder and CEO of Ropedia

An hour of data that links visual observations with motion, spatial structure and task context can offer a richer learning signal than an hour of disconnected footage. Human demonstrations, robot experience and simulation each contribute a different part of that learning process.

4. Where the commercial opportunity lies

Chen sees demand across the whole working life of an intelligent machine.

Which sectors will drive demand

Manufacturing and logistics are natural early sources of demand, because customers there can measure throughput, downtime and the cost of each completed task. Retail operations, facilities services and healthcare support could broaden that demand over time.

For infrastructure providers, Chen argues, the strongest proposition is recurring value tied to better operations. The market only becomes credible when these services earn a continuing place in customers’ operating budgets.

5. How Ropedia is positioning itself

Ropedia describes itself as the infrastructure that turns physical experience into data other companies can use. Its catalogue has four parts:

• HOMIE: the capture hardware
• Foundry: the data processing platform
• Xperience: ready-made datasets
• Physical Data-as-a-Service: collection and delivery built around a customer’s specific needs

Customers can access existing data or work with Ropedia on new collection.

What sets it apart

“Our advantage comes from connecting research with execution,” Chen says. Understanding how models learn tells the company what information to capture and how to judge its usefulness. Building its own capture hardware and processing pipeline gives it control over how that information is produced and delivered.

The company serves robotics and foundation-model companies and has raised US$30 million to support its growth. Its stated aim is to become a long-term infrastructure partner across the ecosystem, judged by how effectively customers can develop and improve their systems.

6. Research as a growth strategy

“Research adoption is crucial because researchers help establish the methods and tools that industry later adopts,” Chen says.

Xperience-10M

Ropedia’s research dataset gives researchers access to a large volume of high-quality real-world data. According to Chen, it has been downloaded more than three million times on Hugging Face.

The Academic Partner Program

The program gives selected researchers access to HOMIE so they can investigate new tasks and environments. Ropedia wants that work to test ideas, reveal limitations and expand what the field can achieve.

The commercial logic

These initiatives help Ropedia understand emerging needs and build relationships with future industry leaders. As research projects move toward deployment, teams may need larger datasets, specialized collection and dependable delivery, and Ropedia wants to be the partner they turn to.

7. Which workflows will benefit first

Chen expects the earliest gains to come from specific workflows, where the task is clear and improvements are easy to measure, rather than from whole industries.

The commercial benefit comes from making automation more adaptable. If a system can cope with changes in packaging or product mix with less engineering work, the economics improve for the operator.

“Across sectors, the decisive question is whether a system can deliver useful work consistently at an acceptable total cost.”

Zhaoxi Chen, Co-Founder and CEO of Ropedia

8. The challenge of scaling profitably

A key challenge, Chen says, is expanding the range of experience Ropedia can provide while keeping quality consistent and costs under control. Collecting more footage does not guarantee a more valuable product. The data has to address something a customer’s model needs to learn.

As robots become more capable, they become candidates for more tasks and environments, which exposes new gaps: different objects, unfamiliar layouts and unusual failures. Progress can therefore create demand for more data even as learning becomes more efficient, and the most valuable data will increasingly target those specific gaps.

How Ropedia is responding

• Standardized capture
• Automated processing
• Quality checks
• Collection priorities guided by customer needs
• The Academic Partner Program, which lowers barriers to research and helps explore new uses

“For our industry, commercial sustainability ultimately depends on delivering useful learning at a cost customers can justify,” Chen says.

9. What investors and enterprise customers should look for

As investment in robotics and embodied AI grows, Chen offers a checklist for judging the companies building the infrastructure behind it.

• Proof of improvement: evidence that the infrastructure makes a customer’s system perform better on relevant tasks, including in unfamiliar conditions
• Traceability: the ability to explain which data contributed to an improvement
• Technical depth: an understanding of how models learn, how physical signals are captured and how quality is measured
• Operational discipline: reliable delivery, clear data rights and quality that holds up as volume grows
• Commercial signals: repeat purchases, and sound economics for producing usable data, including processing and quality control

“Usually dataset size alone cannot answer those questions!” Chen says. These are the standards Ropedia wants to be judged against.

“At the end of the day, a strong infrastructure business earns continued use by making its customers more effective.”

Zhaoxi Chen, Co-Founder and CEO of Ropedia

10. The next three to five years

Chen expects more reusable models and clearer specialization, with companies focusing on robot hardware, intelligence, data infrastructure, or deployment in particular industries. As underlying models improve, more of the commercial work will involve adapting them to specific environments and proving they can operate reliably. Feedback from deployed systems should grow in value, because it shows the gap between a model’s general abilities and the work a customer actually needs done.

Ropedia’s plans

• Expand from data capture and delivery into a broader offering that helps customers adapt models and evaluate performance
• Help establish benchmarks tied to real tasks and outcomes

“Our role should be to make advances in AI easier for the wider robotics industry to turn into useful, economically sustainable deployment.”

Zhaoxi Chen, Co-Founder and CEO of Ropedia

Learn more

More on Ropedia’s datasets, capture hardware and Physical Data-as-a-Service is available at https://ropedia.com.