TraitScout
TraitScout
A robotic phenotyping service for strawberry fields.
Strawberry fields forever.
The Challenge
U.S. strawberry growers are facing a growing labor shortage, driving up operating costs faster than strawberry prices and squeezing already thin margins. Without new solutions, much of the industry could move overseas. L5 Automation aims to prevent this by developing advanced automation.
Introducing TraitScout—our “Physical AI” platform that combines artificial intelligence with robotics to solve real-world, unstructured problems. TraitScout was initially designed for in-field data collection, but is rapidly evolving toward berry harvesting, starting with breeding plots and eventually expanding to commercial farms.
Our main focus is accurate data capture. Using semantic segmentation, we distinguish between plants, beds, and furrows (though we’re not yet identifying blooms or berries). This data helps estimate plant size and shape, which are indicators of harvestability and plant health.
USDA SBIR Phase I: Comprehensive Yield Forecasting
Through this SBIR project, L5 is developing key components for a future system that will deliver much more accurate yield forecasts—especially during the critical 4-6 week window when growers negotiate sales. Better forecasts help growers sell more of their crop in advance at better prices and plan labor needs more precisely, protecting margins and reducing waste.
This project, under USDA SBIR topic 8.13, focuses on improving specialty crop harvesting and developing cyber-physical systems for sustainable agriculture. Our autonomous robots regularly survey strawberry fields, tracking plants over time and counting flowers and berries—essential data for accurate yield predictions.
Building such a system with human labor would be cost-prohibitive, and while drones and rovers have been tried, they can’t count hidden berries because they can’t move foliage safely. L5’s robots, however, can gently move leaves aside to expose the entire plant, enabling accurate counts of all flowers and berries for better forecasting.
Progress and Next Steps
We’ve started with weekly data collection and plan to move to daily as the season progresses and more foliage needs to be moved. Our goal is to ensure consistent, high-quality data throughout all growth stages.
To meet growing field demands, we’re building a second robot dedicated to data collection. The first version will have only cameras, with robotic arms to be added soon. (Suggestions for a name are welcome—our current robot is called Frazor!)
Managing the huge volume of data is a top priority. We’re designing a robust data pipeline for storage, processing, and analytics to extract meaningful insights from the field.
A $175k grant is helping us further develop TraitScout, enabling us to track strawberries throughout their growth cycle.
Achievements
We’re back in the field—data collection for the new season has begun! After initial logistics and testing, we’ve gathered baseline data from new beds. With over a dozen evening visits, we’re collecting about 1.6 TB of video and related data per night across nearly an acre of breeding plots—an order of magnitude more than ever before.
This wealth of data will help breeders better understand their seedling plots and, eventually, all their trials. Already, we’ve made progress in characterizing and extracting information from the beds, using 3D reconstruction and segmentation to identify plants and flowers.
The following video highlights our robot in action, showcasing its ability to navigate the fields, collect data, and perform plant and fruit tracking: