Humyn Labs Secures $47M to Train Robots on Real-World Human Behavior
The robotics training company has closed an early-stage funding round led by Accel, combining equity investment with venture debt to expand operations across emerging markets and build infrastructure for deploying AI-powered robots in household and industrial settings.

Humyn Labs announced a $47 million early-stage funding round aimed at preparing intelligent robots for deployment in real-world conditions across Brazil, South Korea and India. The company converts human movement patterns into training data for robotic systems and AI models.
Accel contributed $26 million in equity funding, joined by Prosus Ventures, BeeGlobal, Finserv Intelligence through Bajaj Finance, Bajaj Alts AI Opportunities Fund, Dallas Venture Capital, Lighthouse Canton and Jump Trading LLC. Lighthouse Canton led the remaining $21 million in venture debt, with additional participation from Trifecta Capital.
Established in early 2026, Humyn operates on the premise that humanoid robot AI systems should acquire dexterity through observation of actual human movement in natural settings rather than staged demonstrations. This approach captures authentic behaviors demonstrating how hands manipulate objects, reach toward targets, apply tactile pressure and adapt when unexpected situations arise.
Physical AI models differ fundamentally from generative AI systems that train on text, images and video. Robots must learn not just visual characteristics and object locations, but also manipulation techniques, force application and error recovery strategies.
"The results can be very, very, very dramatic and catastrophic," co-founder and Chief Executive Manish Agarwal told SiliconANGLE in an interview. Agarwal emphasized that errors tolerable in chatbot applications pose genuine safety risks when robots operate near people, making access to varied, reliable data about human physical interaction essential for training safe, dependable robotic systems.
Humyn constructed a video dataset from head-mounted cameras recording everyday household activities, accumulating up to 200,000 hours of deliberate human motion. The company additionally compiled 120,000 facial images spanning five regions for facial recognition applications, accounting for variations in illumination, angle, expression, eyewear, hair, facial hair, accessories, age and geographic location.
The dataset incorporates more than 50,000 hours of representative speech data for underrepresented language markets, including Bengali, Tamil, Telugu, Odia, Assamese, Korean, Tagalog, Arabic, Vietnamese and Portuguese. While voice AI models already possess extensive data for major languages like English, Mandarin and Spanish, customer support applications in South and Southeast Asian markets remain underserved.
Humyn's platform ingests audio, visual, motion and tactile information, processing these inputs through curation, standardization and annotation procedures that enable AI systems to comprehend motion kinematics, force dynamics and environmental conditions. This integrated approach generates training material for physical AI models, world models, general foundation models, and audio and voice AI systems.
Bringing diversity to robotic data
Geographic diversity matters significantly because robots must grasp the settings and cultural norms of locations where deployment will occur. "Diversity gets you in the room. The quality ensures that you stay in the room," Agarwal said. He underscored how geographic and cultural variation in training data collection shapes robot performance. A household robot trained exclusively in the United States may underperform in an Indian residence, where cooking equipment, appliances, laundry methods and spatial arrangements differ substantially. These distinctions necessitate that robots gain familiarity with the particular environments and behaviors of their eventual users.
By gathering information from more than 20 countries, Humyn trains models on datasets that better match their target deployment scenarios, potentially shortening the timeline to operational readiness.
Agarwal noted that prospective customers include manufacturers of household robots for domestic work and senior care, as well as enterprises pursuing industrial applications involving object handling and navigation through factories, warehouses and commercial spaces.
"The next phase in the deployment cycle is about how well these models perform in the wild," Agarwal said. Humyn's strategy extends beyond data collection alone. The organization aims to document robot failures under actual conditions and leverage those findings to strengthen the AI feedback loop and refine subsequent model iterations. This involves employing computer simulations, physical test units and collaborations with businesses and research institutions to establish field deployment labs that reveal performance discrepancies between predicted outcomes and observed results.
Humyn intends to allocate the new capital toward establishing deployment-readiness infrastructure in Brazil, South Korea and India, advancing its hardware and enrichment engine capabilities, and strengthening its operational footprint in San Francisco, London and Seoul. Agarwal indicated the funding will support testing robots in actual operational environments to uncover performance gaps, with findings channeled back into model refinement processes.


