A deep reinforcement learning-based multi-agent area coverage control for smart agriculture

Precision agriculture (PA) is a collage of strategies and technologies to optimize operations and decisions in farms by using spatial and temporal variabilities in yield, crops, and soil within an agricultural plot. It is a data-driven technique, therefore, selective treatment of crops and soil, and managing variabilities using robots and smart sensors is the next improvement in PA. In this paper, it is modeled as a multi-agent patrolling problem, where robots visit subregions that required immediate attention in the agricultural field. Furthermore, for area coverage / patrolling task in the agricultural plot, a centralized Convolutional Neural Network (CNN) based Dual Deep Q-learning (DDQN) is proposed. A customized reward function is designed, which rewards worth-visiting idle regions, and punishes undesirable actions. A proposed algorithm has been compared with various algorithms including individual Q-learning (IRL), uniform coverage (UC), and Behavior-Based Robotics coverage (BBR) for different scenarios in the agricultural plots.