AI digital twin estimates lettuce biomass within 1.5g in hydroponics
New digital twin predicts lettuce yield with 2g accuracy up to 4 days out.
Researchers from Carnegie Mellon University (Morgan Mayborne, Abhisesh Silwal, George Kantor) have developed a measurement-driven digital twin architecture that tracks individual lettuce plants in hydroponic greenhouses. The system combines custom sensor hardware, a neural network for real-time mass estimation, and growth models to continuously update yield forecasts. The neural network, trained on 1,300 RGB-D images, estimates plant biomass within 1.5 grams of ground truth. Once integrated, the digital twin predicts future yield one to four days ahead with a forecasting error of approximately 2 grams.
This approach addresses a critical need in urban hydroponics: reliable, plant-level growth monitoring without destructive sampling. By fusing continuous sensor data with model-based predictions, the digital twin adapts to each plant's unique growth trajectory. The result is a robust framework that could be extended to other crops, enabling data-driven decisions for harvest timing, resource allocation, and supply chain planning. The paper was published on arXiv (cs.RO, June 2026) and represents a practical step toward AI-powered precision agriculture.
- Custom neural network uses 1,300 RGB-D images to estimate plant mass within 1.5g accuracy.
- Digital twin forecasts lettuce yield 1–4 days ahead with a ~2g error margin.
- System integrates sensor hardware, machine learning, and growth models for real-time tracking in hydroponic greenhouses.
Why It Matters
Enables precise, non-destructive yield forecasting in urban hydroponics, optimizing harvest timing and resource use.