Singapore-based robotics data infrastructure firm Ropedia Pte. Ltd. today announced the launch of HOMIE Gen2, the next generation of the company’s head-mounted wearable that records human movement data to train robots.
Smart robotics requires tremendous amounts of high-quality data taken from various environments. Most of the training data used to build the artificial intelligence foundation models that drive robots comes from videos of humans doing tasks such as picking up objects, moving them and completing other simple or procedural tasks.
Not all of this video content exists in forms that can be easily transformed into useful data for training robots.
“Scaling laws make large language models work because more data equals better results,” said co-founder and Chief Executive Zhaoxi Chen. “Physical AI runs on the same law, except the data needs to come from someone actually doing the job. For example, imagine learning how to drive by only sitting in the back seat.”
AI models benefit greatly not just from volumes of high-quality data, but also from the right type of data. Ropedia’s wearable capture device puts AI models “in the driver’s seat,” rather than at an angle, a distance or from the side. The idea is to provide video of activities that robotic arms will mimic from the point of view of cameras positioned where they will be in most robotic setups, especially humanoid and arm-based modeling.
The company said HOMIE Gen2 is designed to be the first egocentric capture device with a 360-degree field of view for robot training. It treats the human experience for robots as immersive, location- and environment-specific training that AI models can translate into virtual training, which can then be used to mimic physical actions.
Multiple cameras for immersive capture
The Gen2 uses four camera streams, spatial audio and motion data to capture human activity as the wearer perceives it. The device itself is shaped like a crown that sits atop the head and moves with head movements, watching hands and body, listening to the environment, and recording the space and how the user’s movements relate to it – not just recording footage.
As recordings carry video, camera pose, depth, hand and body key points and task annotations, it provides rich metadata for robotics foundation models to quickly get up to speed with training. A single session with the Gen2 can train for several target tasks or reconstruct a single episode of human behavior.
“Hardware-synchronizing four camera streams, spatial audio and inertial data to within 50 microseconds means our models can treat a session as one continuous signal instead of reconciling four separate ones,” said Chief Scientist Ziwei Liu, who’s an associate professor at Nanyang Technological University in Singapore.
The company added that the wearable requires no external rig, allowing users to capture runs across any environment. This includes homes, factories or shopping malls – without elaborate setup or an engineering team. Deployment is 10 times faster than the previous generation and costs about one-12th the cost of a full lab capture.
Ropedia did not disclose pricing but said business users can purchase the devices individually or in bundles. The company is open to partnerships, pilot programs and direct sales.
The release of the HOMIE Gen2 follows a $22 million pre-Series A funding round raised last month to scale up its engineering efforts.