Humyn Labs raises $47M to train robots on real-world human motion
The startup aims to bridge the gap between AI training and physical robotics by capturing how humans interact with objects in diverse global environments.

The short version
- Humyn Labs raised $47 million in a funding round led by Accel and Lighthouse Canton.
- The company captured 200,000 hours of intentional human motion using head-mounted cameras.
- The dataset includes 120,000 faces and 50,000 hours of speech for underserved language markets.
- The startup is expanding its presence to Brazil, South Korea, and India to capture geographic diversity.
Why it matters
Current AI models often struggle with the physics of the real world, such as how much force to apply to an object. By providing robots with data on human mistakes and successes, Humyn Labs helps make autonomous machines safer for use in homes and factories.
Bridging the physical gap
Humyn Labs recently announced it raised $47 million in early-stage funding to accelerate the deployment of intelligent robots. The company focuses on turning human experience into training data for artificial intelligence and robotics. They believe that for humanoid robots to function, they must learn from observing how humans actually move in real environments.
Unlike standard generative AI models that learn from text or images, physical AI models require an understanding of the physical world. This includes knowing how to reach for, touch, handle, and recover from errors when interacting with objects. Humyn Labs argues that robots cannot learn these complex dexterities from staged demonstrations alone.
The funding round was led by Accel, which provided $26 million in equity. Other participants included Prosus Ventures, BeeGlobal, and Jump Trading LLC. Lighthouse Canton led a separate $21 million venture debt portion of the round with participation from Trifecta Capital.
Capturing human motion
To build its training library, Humyn Labs created a video corpus using head-mounted cameras. Participants performed everyday household tasks, resulting in up to 200,000 hours of intentional human motion. This data allows the AI to see how hands work on objects without needing explicit instructions from a programmer.
The company's engine processes four primary inputs: sound, sight, motion, and touch. It curates, normalizes, and labels this information so that AI models can understand the kinematics of motion and environmental factors. This process helps create training patterns for world models and general foundation models.
Beyond physical motion, the company also shipped a dataset of 120,000 faces across five regions for facial recognition. They also included more than 50,000 hours of speech data for underserved markets, such as Bengali, Tagalog, and Vietnamese. This ensures that the AI understands diverse voices and faces that are often missing from standard datasets.
The importance of diversity
Geographic diversity is a core component of Humyn Labs' strategy for robot training. A robot trained only in the United States might struggle in an Indian home because of differences in kitchen appliances, laundry practices, and room layouts. The company is collecting data across more than 20 countries to ensure robots can adapt to different cultural practices.
Manish Agarwal, the company's co-founder and CEO, noted that diversity is what gets a robot into a room, while quality is what keeps it there. He emphasized that robots must understand the specific environments and practices of the people they are intended to serve. This approach aims to make robots more reliable in varied global settings.
The company is currently expanding its on-the-ground presence to San Francisco, London, and Seoul. They are also building deployment-readiness infrastructure in Brazil, South Korea, and India. These locations will serve as hubs for collecting the high-quality, diverse data needed for real-world training.
Safety and future testing
One of the primary goals of this data collection is to improve robot safety. Agarwal told SiliconANGLE that mistakes for a robot can be "very, very, very dramatic and catastrophic" compared to a chatbot. Because robots operate alongside people, they must be trained to handle unexpected situations without causing harm.
The company plans to use the new funding to develop its hardware and enrichment engine. They also intend to use virtual simulations and physical replicas to test how models perform. By identifying where robots fail in the wild, they can feed those results back into the AI flywheel to improve future versions.
Current interest in Humyn Labs' technology comes from several sectors. Customers are looking to build household robots for domestic tasks and elder care, as well as industrial robots for factories and warehouses. The next phase of the company's cycle focuses on how well these models perform in unpredictable, real-world conditions.
What to watch
- How the company handles the privacy implications of head-mounted camera data.
- Which specific household tasks will be prioritized for the first wave of domestic robots.
Try it
Look up the different kitchen layouts and cooking tools used in India versus the United States to see how complex a robot's task environment can be.
Sources and links
- ReportingSiliconANGLE: the article we sourced
First reported by SiliconANGLE. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.