Ai-powered soybean yield prediction using deep learning and ground robotics
An autonomous phenotyping platform that combines deep-learning algorithms with a robotic ground rover to predict soybean yield components (pod count, seeds per pod, seed weight) in-field without harvesting. The system delivers high-throughput, non-destructive yield estimation to accelerate plant breeding programs.
Overview
This solution offers a non-destructive approach to predicting soybean yield by combining deep-learning computer vision algorithms with an autonomous ground robotic platform. Instead of harvesting experimental plots to measure yield, the system analyzes images captured in-field to determine key yield components, including pod number, seeds per pod, and seed weight. By digitizing and automating data collection, the platform helps seed companies and plant breeding programs accelerate line selection, reduce labor costs, and standardize phenotypic evaluation across large trial networks.
Technical specifications
Core capabilities:
- Deep-learning algorithms trained to detect and count soybean pods, determine seed counts, and measure pod dimensions from images
- Ground rover platform controlled by the Robotic Operating System (ROS), capable of navigating soybean plots and capturing standardized image data
- Pod-counting algorithm validated with a Pearson correlation of r=0.93 against manual counts, achieving approximately 90% precision across 100 plots with varying plant densities (5-35 plants per meter) at full maturity (R8 stage)
- Web-based platform paired with a mobile application that allows field technicians to record and monitor phenotypic data in real time
- Designed for integration into existing breeding workflows, providing data-driven insights on yield potential and line performance without destructive sampling
Technology readiness level
The pod-counting algorithm has been validated under controlled conditions, and the robotic platform has been field-tested at an Argentinian plant breeding center during the 2020-2021 soybean crop season. The next validation phase, planned over approximately six months, will assess the algorithms' ability to predict yield components under real field conditions and to recognize high-yielding soybean lines using a new ROS-controlled data-collection platform. The team is actively seeking seed-company partners to support field trials, data collection, and harvest comparisons.
About GBOT SOLUTIONS
GBOT Solutions is an agricultural technology company that specializes in developing autonomous terrestrial robots, or rovers, designed for high-throughput crop phenotyping and precision data collection. The platform integrates robotics with artificial intelligence, using sensors, cameras, and deep learning algorithms to analyze phenotypic traits such as crop yield potential and disease resistance. By automating traditional, labor-intensive evaluation processes, the company provides plant breeding programs and agricultural firms with detailed, data-driven insights to accelerate the development of improved, more resilient crop varieties. Additionally, the company is researching sustainable applications for its technology, including the use of precision mapping and laser-based weed control to reduce reliance on traditional agrochemicals.
This technology provides significant value to the agricultural sector by digitizing and standardizing field data collection, which enhances the efficiency of genetic research and crop management. The company operates a web-based platform supported by a mobile application that allows field technicians to record and monitor phenotypic data in real-time. By moving away from manual, error-prone assessment methods toward autonomous, sensor-led monitoring, GBOT assists breeders and producers in making informed decisions about plant selection and resource allocation. Headquartered in Sunchales, Argentina, the company serves as a partner for organizations aiming to modernize agricultural practices through innovation and sustainable automation.
delivered to your inbox.