A Multi-Agent Reinforcement Learning Approach for Spatiotemporal Sensing Application in Precision Agriculture

Digital transformations within the realm of Industry 4.0 have introduced a paradigm shift in the management/production systems of small- and medium-sized enterprises (SMEs) in the agriculture sector. Embracing such ideas as "smart farming" and "precision agriculture", farmers and agricultural SMEs are now using wireless sensor networks and autonomous mobile robots (e.g. wheeled mobile robots and drones) to help maximise their production output while at the same time minimising their farming cost. One characteristic of such networks of sensors and mobile robots is that the individual sensor or mobile robot in the network has only limited sensing or movement coverage abilities, respectively. In this regard, an important challenge in their implementation is how to coordinate each individual sensor (or mobile robot) which has partial sensing (or movement coverage) ability with the other to ensure that their resulting network is capable of providing full area sensing (or movement coverage) of the farming field. This chapter discusses recent trends in combining sensor/actuator networks and machine learning techniques (i.e. reinforcement learning and the kernel-based method) for spatiotemporal modelling and control purposes in smart farming applications. Illustrative examples and the advantages of their use in agricultural SMEs are also provided.