A model to represent human activities in farming systems based on reactive situated agents

This paper proposes a model to represent and simulate human activities in farming systems. Our definition of the activity concept (otherwise said ‘action’) stems from the ‘situated action’ theory (Suchman, 1987). Action may be summarized simply as ‘what people are actually doing’. Therefore, we characterize action according to four dimensions: actors (persons or machines engaged in activity), time (start date, end date and duration), space (locations where action occurs) and the interaction with the actor’s environment. To take into account all these dimensions our model is based on: (i) an ‘agent’ approach, (ii) a mathematical formalism to represent action over time (Guerrin, 2009) and (iii) a discrete representation of space inspired from cellular automata. Our aim is to apply this model to representing the human component of agricultural systems, i.e. the farming practices. Space is modelled as a regular grid, each cell being controlled by a specific autonomous abstract ‘agent’, in the sense of multiagent systems (MAS), called ‘cell-agent’. Since activity depends, according to the situated action paradigm, on its context of occurrence and continuously involves interactions with the environment, we focused on the concepts of stigmergy and affordance to carry the information necessary to realize actions and on a reactive agent approach. The radical notion of reactive agent postulates that agents’ actions are reac-tions to stimuli. Various constraints to be dealt with are imposed by our application domain (farming sys-tems), namely temporal constraints the cell-agents must consider for acting. Consequently, our model, alt-hough mainly based on a stimulus-response mechanism, involves also a process, kept as simple as possible, allowing spatially situated agents to select the most appropriate action when several actions are concurrently possible. For this, priority levels coupled with the stigmergy and affordance concepts are used. Cell-agent actions are realized through specific objects called ‘actuators’ representing, unconventionally, the real actors (e.g. a farmer possibly using equipments, animals, etc.). An actuator is a non-autonomous entity that receives from the cell-agent where it is located the command of actions it must execute. A cell-agent can perceive, at any time, what happens in its neighbouring cells and which are the objects these cells host. In case a cell-agent needs a resource not located in a nearby cell, it can possibly get it by propagating its request through its neighbours. Action is represented by a binary function of time (Guerrin, 2009) and implemented by these ‘actuators’, exhibiting dynamic behaviours, although they are actually directed by the cell-agents they rely on. Action starts and ends according to events detected on processes observed by the cell-agents in their local environ-ment, provided they can access the necessary resources, among which actuators endowed with relevant abili-ties. The course of action is subject to a set of conditions that must remain satisfied for the action goes to its end. In the meanwhile, it may be interrupted, resumed or cancelled. The originality of this work lies in that real ‘actors’ (like persons, animals, machines, etc.) are represented as non-autonomous ‘objects’ and not as autonomous ‘agents’, as it is generally the case in other modelling efforts (e.g., Sierhuis et al., 2006). These objects cannot act by themselves. They just follow the instructions given by the cell-agents they rely upon. Our model allows so one to represent the behaviour of real ‘situated’ actors as if it were the environment that dictated them what to do. Doing this way complies with the situated action theory.