Realset AI and Flatkey Raise $10M Series A for Real-World AI Training Data
Realset AI and Flatkey have secured $10 million in Series A funding to expand the capture of real-world data for training AI models and embodied agents. The investment aims to bridge the gap between simulated and physical task performance.
San Jose, Calif. – September 23, 2026 – AI companies Realset AI and Flatkey announced today they have raised $10 million in Series A funding. The capital will be used to expand Realset AI's network of real-world capture environments, including workplaces and studios, and to grow its pool of expert human demonstrators and domain experts. The goal is to produce training data crucial for advanced AI models and robotics.
The companies highlight that readily available text and image data from the internet is becoming exhausted. They assert that the next significant advancements in AI will come from data that captures humans performing tasks in physical environments, such as household chores or warehouse operations. This type of data is not available online, and simulations struggle to accurately replicate it.
Realset AI's approach involves recording real people performing tasks in authentic settings. The company utilizes expert demonstrators rather than crowd annotators and produces de-identified data. This data is delivered from cloud infrastructure hosted in the U.S., with continuous quality monitoring throughout the pipeline.
The captured data is categorized into three offerings: Realset Body for training physical agents, Realset Field for creating virtual environments from real workflows, and Realset Judge for evaluating and monitoring AI agents in production. All data undergoes rigorous quality checks and is verifiable.
Realset AI also plans to release open benchmarks to measure AI model performance in real-world tasks outside laboratory settings. The first benchmark, focusing on household manipulation tasks, is expected in the fourth quarter of 2026.