Company & Market · Robot24.com Original

Realset AI, Flatkey Raise $10M to Expand Real-World Robot Training Data

Realset is collecting data from people performing physical tasks for use in training AI models and embodied agents.

September 25, 2026 by Artem

Realset graphic announcing $10 million in Series A funding for real-world data used to train frontier AI models and embodied agents.

SAN JOSE, Calif. — Realset AI and Flatkey raised $10 million in a Series A funding round, according to Flatkey, as the companies develop businesses around AI training data and infrastructure.

The two companies had raised $10 million in Series A funding, Flatkey said in a Sept. 24 announcement. The release did not name investors, give a valuation, or say how the $10 million will be split between the two firms.
Realset is gathering data from people doing physical work to train AI models and embodied agents. Examples include folding laundry, loading a dishwasher and packing orders. Its Realset Body service also collects bimanual teleoperation sessions alongside IMU readings and robot action logs.

Those datasets support development of vision–language–action models that integrate visual perception and language instructions to guide physical actions. The models assist robots with manipulation tasks instead of simply executing predefined motions.

Realset collects its data from expert demonstrators rather than relying solely on crowdsourced annotation, according to the company. Its broader business also includes reinforcement-learning environments modeled on workflows in areas such as logistics and manufacturing, along with human evaluation of AI agents. Approved data records include provenance information and undergo an independent quality review.

Flatkey provides developers with access to multiple AI models and software tools through a single application programming interface, or API, key. The company said it will use funding from the Series A to add models and tools and expand the infrastructure that routes developer traffic to them. Flatkey also said more than 10,000 developers had signed up for its platform since the platform launched in July.

Realset intends to release its household‑manipulation benchmark results in Q4 2026 and says the benchmark will evaluate open‑source robot policies (π0, OpenVLA, Nvidia's GR00T and Octo) on folding, loading, sorting and wiping.

Realset and Flatkey have not disclosed Series A investors, valuation, or how the $10 million will be divided.