The collection of micro-climatic information through a mobile robot

Abstract

Temperature fluctuations and, in general, climatic conditions can significantly affect the chemical composition of grapes and, in turn, the taste and aromas of wine. At the same time, having a complete climatic mapping of the vineyard is a challenging topic. Traditional solutions are based on using sensors displaced in strategic points of interest. However, in many cases, disseminating a large number of fixed sensors capable of monitoring environmental parameters in a capillary manner is impractical in terms of costs and space constraints. For this reason, some alternative solutions have been envisioned by combining a limited number of sensors and artificial intelligence solutions for building precise predictive models. Our recent investigation explored the possibility of exploiting an agricultural robot performing repetitive tasks, such as weeding, to collect capillary information during its journey. The low cost of equipping existing agricultural robots with dedicated sensors and the technological advances in artificial intelligence constitute the enabling factors to be explored and exploited in this application area. In this research, we aimed to make the path of the agricultural robot “expert” to best combine its operational activities with data acquisition and energy recharge. In particular, the Reinforcement Learning technique was used to learn from the past and identify the best trajectory and stopping pattern the robot should follow in a specific contextual situation. At the same time, convolutional and recurrent networks are applied to discover the correlations, respectively spatial and temporal, in the values of the monitored environmental parameters and further guide the robot’s decisions. In summary, the use of mobile sensors accompanied by advanced artificial intelligence techniques can be a valid tool to create a “digital twin” of the vineyard to monitor and estimate the trend of specific climatic quantities, such as temperature and humidity, but also, in the future, the phenological stage of each plant and the development of pathogens. The combination of all these tools can lead to a next generation of decision support systems for viticulture, which help not only to prevent plant diseases, predict and optimize their growth, but also optimize the usage of resources, like energy, water, and pesticides, to increase the quality of the product while reducing the waste of precious resources.