AI Startup HUMXN Offers Free Home Repairs to Collect Robotics Training Data
AI startup HUMXN is subsidizing home repair and maintenance services in exchange for permission to record technicians at work. The company aims to collect real-world training data from residential environments to help train next-generation AI robots. As AI development faces a major physical bottleneck in acquiring high-quality real-world data, this initiative offers a novel consent-based economic model to capture complex physical labor. By paying for real-world scenarios rather than scraping web data, it bridges the gap between human blue-collar expertise and embodied AI training. The project is initially launching across three U.S. cities—Minneapolis, Chicago, and Miami—to gather diverse environmental and weather conditions. While homeowners receive free services and technicians get paid normally, HUMXN operators wear data-collecting devices to track the entire troubleshooting and repair process.
## BACKGROUND
Embodied AI refers to AI systems embedded within a physical form, such as a robot, that perceive and act upon the physical world. Training these systems requires multimodal episodes of real-world sensor inputs and motor actions, making the acquisition of diverse physical training data a critical challenge in robotics.