Optimizing Trajectories for Rechargeable Agricultural Robots in Greenhouse Climatic Sensing Using Deep Reinforcement Learning with Proximal Policy Optimization Algorithm ^†

The experimentation of agricultural robots has been increasing in recent years, both in greenhouses and open fields. While agricultural robots are inherently useful for automating various farming tasks, their presence can also be leveraged to collect measurements along their paths. This approach enables the creation of a complete and detailed picture of the climate conditions inside a greenhouse, reducing the need to distribute a large number of physical devices among the crops. In this regard, choosing the best visiting sequence of the Points of Interest (PoIs) regarding where to perform the measurements deserves particular attention. This trajectory planning has to carefully combine the amount and significance of the collected data with the energy requirements of the robot. In this paper, we propose a method based on Deep Reinforcement Learning enriched with a Proximal Policy Optimization (PPO) algorithm for determining the best trajectory an agricultural robot must follow to balance the number of measurements and autonomy adequately. The proposed approach has been applied to a real-world case study regarding a greenhouse in Verona (Italy) and compared with other existing state-of-the-art approaches.