Multiple aerial/ground vehicles coordinated spraying using reinforcement learning

Investments in unmanned aerial vehicles (UAVs) have recently surged in precision agriculture. However, multi-UAV missions can face limitations due to weather conditions, highlighting the need for effective spray coverage. A novel system tailored for spraying in windy conditions to tackle this challenge is proposed. Instead of directly controlling sprayed drops, the location of spraying UAVs based on real-time wind data is adjusted. Our proposed methodology consists of three stages: Firstly, on-policy reinforcement learning (RL) with Proximal Policy Optimization (PPO) is utilized to optimize path planning. In the second stage, another PPO iteration to correct wind drift is employed, leveraging the latest wind data to enhance spray mission efficiency. Lastly, a novel algorithm is introduced to improve efficiency in narrow areas by substituting unmanned aerial vehicles with unmanned ground vehicles. To evaluate the efficiency of the proposed aerial spraying system, we conducted a simulation and reported the corresponding results.