Autonomous UAV Navigation and Mapping for Accurate Fruit Detection and Counting in Controlled Environments: Simulation and Real-World Validation

This paper presents a solution for high accuracy fruit counting in a Controlled Environment Agriculture (CEA) settings using Unmanned Aerial Vehicles (UAV) implemented in simulation and real-world scenarios. Using the fine-tuned YOLOv8 [1] model for precise object detection and classification along with a custom path planning algorithm for simulation. This approach integrates a hybrid A* Traveling Salesman Problem (TSP) algorithm for efficient 3D path planning. The solution is further extended to the real-life scenario using LiDAR based mapping and point-cloud filtering techniques to avoid recounting of fruits. The suggested methodology increases operating efficiency, reduces dependency on human labor, and improves accuracy. Experimental results, derived from both simulations and real-world testing, achieving a fruitcount accuracy of 98 % and 90 % respectively, demonstrate the effectiveness of this integrated approach. This solution was implemented in the International Conference of Unmanned Aircraft System (ICUAS) [2] UAV Competition 2024, securing <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$2^{\text {nd }}$</tex> position in the simulation phase (out of 24 teams) and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$3^{\text {rd }}$</tex> overall in the real-world phase.