AI trained in virtual farms improves fruit detection and reduces costs
A research team from the University of Florida has developed a digital twin—a detailed computer model—of a strawberry farm to enhance agricultural robotics and artificial intelligence tools. This model enables testing year-round, regardless of Florida’s seasonal strawberry production.
Digital twins allow virtual simulation of real-world environments. In this case, researchers created a life-sized virtual strawberry field to train AI systems.
The team, led by assistant professor Dana Choi, used this digital farm to test robotic fruit detection with no real-world data.
The AI model achieved 92% accuracy in spotting strawberries and estimated their diameter with only 1.2 millimetres of error. These results are considered accurate enough for commercial grading and decision-making on harvest timing.
“Because the computer-simulated field never goes out of season, new berry-spotting tools can be prototyped even in the summer – speeding innovation,” said Choi.
Traditionally, training such systems required thousands of real images, each manually labelled—a time-consuming and expensive process. The digital twin eliminates that need, allowing synthetic images to be created and labelled instantly.
Beyond fruit detection, this technology supports development of smart sprayers, robotic pickers, and operator training—all without field access. It significantly reduces costs and boosts efficiency in agriculture technology innovation.
Florida’s strawberry industry is worth $500 million, with national production reaching $2 billion. This breakthrough helps reduce production costs while addressing labour challenges and improving overall efficiency.
With digital twin tools, researchers can now bring robotic solutions to market faster, helping farmers use advanced tools at lower costs and with greater precision.