Hungarian startup DiffuseDrive generates synthetic training data for physical AI systems, focusing on sectors like defence, aerospace, mining, and autonomous vehicles. The platform identifies gaps in existing customer datasets and creates targeted synthetic data to improve model performance in rare or dangerous real-world scenarios.
The company was founded by former employees of the Bosch vehicle automation division. Its technology allows for precise control over object placement and size within generated images, which is critical for training perception systems to recognize distant or small objects in various fields of view.
DiffuseDrive reports that integrating its synthetic data with real-world datasets can improve model performance by more than 10 percent. The system can be deployed on customer infrastructure, including air-gapped environments, and can generate new training data in a matter of hours depending on available GPU resources.
In 2025, the startup raised $3.5 million in seed funding. The company is currently expanding its focus toward the robotics sector, aiming to produce dense annotations for robotic experiments, including sensor streams and state changes, to better interpret complex operational data.
The leadership team includes VP of AI Engineering Gabor Vecsei and AI Research Lead Daniel Schmid. Looking ahead, the company anticipates a market shift toward smaller, specialized AI models to optimize compute costs and performance, particularly within the European technology landscape.
