Humyn Labs 获 4700 万美元 A 轮融资用于采集现实世界机器人训练数据
Humyn Labs Secures $47M Series A for Real-World Robotics Data
印度 Humyn Labs 完成 4700 万美元 A 轮融资,为 AI 和机器人模型提供物理环境训练数据。该公司使用自有双摄像头硬件在巴西、印度等 20 多个国家采集日常任务视频,经软件处理后输出 MCAP、RLDS、LeRobot v3 等格式。创始人称已实现 1000 万美元营收并签下 2000 万美元合同,主要客户来自美国。
主题
正文
TechCrunch reported on November 18, 2026, that Bengaluru-based Humyn Labs had raised $47 million in Series A financing for its business supplying data used to train AI and robotics models that operate in physical environments.
The financing comprises $26 million in equity led by Accel and $21 million in venture debt. Prosus Ventures, BeeGlobal, Bajaj Finance’s Finserv Intelligence, Bajaj Alts AI Opportunities Fund, Dallas Venture Capital, Lighthouse Canton and Jump Trading participated in the equity component. Lighthouse Canton Growth Debt Fund II led the debt, with Trifecta Capital also participating.
Manish Agarwal, former CEO of Indian gaming company Nazara Technologies, and Ishank Gupta, a former AB InBev executive, founded Humyn in February 2026. They previously founded Kratos Studios, now known as KGeN. The founders said KGeN’s contributor network across emerging markets helps Humyn collect recordings inside people’s homes.
Agarwal said growth has increased Humyn’s working-capital needs because it pays workers and collection partners daily or weekly while typically waiting 30 to 60 days for AI-lab customers to pay, with borrowing bridging that gap and causing less dilution than a further equity sale.
The company uses its own two-camera hardware to record people performing everyday tasks. Agarwal said its software processes and labels the footage, tracking hand movements, interactions with objects and other details to turn raw recordings into data from which AI models can learn.
Collection extends across more than 20 countries in Latin America and Asia, with Brazil, India and South Korea as the main hubs, Agarwal said. Rather than relying on staged activities, Humyn records people doing their usual work or household chores, while local partners identify sites and supervise the process.
Humyn’s platform description sets out a broader pipeline spanning sound, sight, motion and touch. Speech audio, first-person video, motion traces from inertial measurement units and grip-force data pass through validation, processing, labelling and human review before being combined into a synchronized training signal. Listed outputs include MCAP, RLDS, LeRobot v3, ROS Bag, WAV, FLAC and JSON.
Reported Demand and Contracted Work
Gupta said Humyn has 12 paying customers spanning frontier AI labs, robotics companies and developers of world models, and that about 90% of its revenue comes from the United States. He declined to name specific customers.
Agarwal reported $10 million in revenue between April and October 2026, another $20 million in signed contracts, and a further $20 million in deals that have not been finalized. The last amount is prospective business rather than part of the signed-contract total.
Agarwal said the company has 51 full-time employees, most based in Bengaluru, alongside local partners across its collection countries.
The reported revenue and contracts point to early demand, while the payment cycle shows the financial demands of meeting it. Humyn’s central operating constraint is to sustain a distributed collection network and its processing pipeline while covering the delay between paying for recordings and receiving customer revenue.
AI 热点