VINES: Track 1: Enabling Real-Time Agricultural Robotics with AI-Native Cellular Access Edge

Modern farms increasingly rely on robots to perform delicate tasks such as picking strawberries quickly without bruising the fruit. These robots need powerful onboard computers to perform tasks such as seeing and planning. However, such computers are heavy, costly, and power-hungry, which makes the robots large and expensive to deploy widely. This project is developing a system called FarmBrain that keeps simple, lightweight, and low-cost robots operational by offloading demanding artificial intelligence (AI) tasks, such as seeing and planning, in real time to a nearby, powerful computer on the network edge via a private cellular wireless network. Each robot is equipped with fast safety reactions, and many robots share a single computer, lowering the overall cost of automation. The project is organized into three connected thrusts. The first thrust develops a programming model that allows developers to express a manipulation task as cooperating control loops running at different rates. Each loop has explicit service-level objectives and a unified way to recover from robot and network failures. The second thrust builds a compiler and runtime that automatically splits these loops between robots and edge servers. Then, it places, schedules, and elastically scales them to meet strict deadlines under changing wireless conditions and workloads. The third thrust creates an AI-based management system that uses a graph neural network to diagnose performance problems spanning the robot, the network, and the edge and to maintain radio access network availability during failures. By lowering the barrier to automation for small and medium farms, this project makes agricultural robotics economically practical and helps address farm labor shortages while supporting food production and reducing the use of water, fertilizer, and pesticides through more precise plant care. The underlying approach to real-time computing at the network edge can also be applied to other safety-critical settings, such as autonomous vehicles, industrial automation, and emergency response. The project releases its software, tools, example implementations, and benchmarks as open source. It also integrates its findings into computer systems and robotics courses. The project mentors students through a research initiative on agricultural robotics and engages industry partners and local farmers to keep the work grounded in real needs. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.