Ropedia Launches HOMIE Gen2 to Scale Real-World Data for Physical AI

SINGAPORE--(BUSINESS WIRE)--Ropedia, a Singapore-based company building data infrastructure for physical AI, today launched HOMIE Gen2, a head-mounted capture system that records the wearer’s environment and converts it into structured, training-ready data for robot learning.

The system is the fourth generation of Ropedia’s capture hardware in 12 months, and follows the company’s recent pre-Series A round which brought total funding to US$30 million.

As Physical AI develops, researchers and robotics companies are increasingly looking to real-world experience as a source of training data. Unlike digital AI systems that can learn from large collections of text and images, robots need data that captures how people perceive, move and interact with physical environments. With HOMIE Gen2, Ropedia is building a scalable engine that transforms human experience into structured and scalable data for robot learning, simulation and embodied AI.

“LLMs have shown that scaling training data can unlock new levels of intelligence. Physical AI follows a similar scaling principle, but its training data must capture the complexity of the real world. Traditional fixed-camera setups and controlled demonstrations can capture individual actions, but they struggle to provide the diversity and in-the-wild data needed to train systems for the physical world. As human experience scales in volume, diversity and quality, it can provide a broader foundation for Physical AI to learn how people perceive and act in real environments,” said Zhaoxi Chen, chief executive and co-founder of Ropedia.

Built for Real-World Capture

HOMIE Gen2 is designed for wearable, in-the-wild data collection. The system provides 360-degree visual coverage, 1080p video at 30FPS with global shutter and four-channel spatial audio in a lightweight, ergonomic wearable weighing just 380 grams. A multifunction port supports hardware synchronization across devices and future sensing expansion, allowing multiple HOMIE Gen2 units to capture the same environment from different people and perspectives.

The self-contained design enables hands-free capture across homes, factories, retail environments and other real-world settings, without requiring external cameras or dedicated capture infrastructure. This allows everyday activities and work tasks to be recorded in their natural environments rather than recreated specifically for robot training.

“Essentially, HOMIE acts as an enabler for our Human Experience Engine. It is the infrastructure layer between the physical world and foundation models that converts human activity into structured data for robot learning. Most robots are trained as if they were learning to drive from the back seat. HOMIE Gen2 puts them behind the wheel,” said Chen.

From Capture to Training-Ready Data

The captured data is processed through Ropedia’s Human Experience Engine, where more than 10 modalities of human experience — including visual, audio, depth, motion, pose and spatial signals — are synchronized, processed and structured into training-ready data. Testing has demonstrated up to 96% accuracy in capturing and understanding real-world human activity.

By turning raw human experience into structured data with synchronized multimodal context, Ropedia aims to reduce the downstream effort required to prepare data for Physical AI, helping robotics teams move more quickly from real-world capture to model training.

“We are at a stage for Physical AI similar to where autonomous driving was before widespread deployment,” added Chen. “Autonomous driving has millions of vehicles on the road generating real-world data. Physical AI doesn’t have that scale of robot deployment yet. Human data helps bridge that gap, giving us the real-world experience needed to move the industry forward.”

HOMIE was one of the first dedicated egocentric 3D capture systems developed by Ropedia, enabling the company to build high-quality human experience data at scale. This foundation powered Xperience-10M, Ropedia’s flagship human experience dataset, released in March 2026 with more than 10 million interaction episodes and 10,000 hours of first-person recordings.

The launch of HOMIE Gen2 builds on that foundation with improvements in camera quality, sensing fidelity, multimodal coverage and hardware synchronization, expanding Ropedia’s ability to capture and process human experience for Physical AI training.

About Ropedia

Ropedia builds data infrastructure for physical AI. Its platform combines HOMIE, a human-centric multimodal capture system; Experience Engine, the infrastructure for processing and structuring human experience; and Xperience datasets and data services for robotics and embodied AI teams.

Ropedia was founded in Singapore in the second half of 2025 by Zhaoxi Chen, Fangzhou Hong and Ziwei Liu. The company is headquartered in Singapore, with offices in Mountain View, California; Kuala Lumpur; and Shanghai.

Additional resources: sample dataset and dataset card on Hugging Face; full Xperience-10M available on Hugging Face under controlled access for approved non-commercial research; HOMIE-toolkit on GitHub; full HOMIE Gen2 technical documentation: https://ropedia.com/homie (will be available after embargo).