The University of Missouri introduced FieldVision in a 23 September 2026 research report. The framework uses multi-agent reinforcement learning to decide where agricultural drone imagery should be processed: onboard an aircraft, at a nearby server or in the cloud. Its training accounts for competition over shared computing and communications resources.
After centralised training, individual drones make decisions from locally available information without continuous communication between aircraft. The reported comparisons were simulation tests, where FieldVision improved selected performance and reliability measures against rule-based and single-drone approaches. Those findings support further investigation but do not demonstrate field performance under actual rural network outages, aircraft energy constraints or changing crop conditions.
