Constrasting the standard framework based on agent-to-agent direct communication, in this paper we focus on agent interaction with the environment. In particular, the environment is seen as populated by
, automatising coordination tasks and mediating agent interaction.
As a semantic framework for this context we propose operating instructions, which the agent reads and interprets to become aware of the allowed actions and perceptions at a given time, and the MAS infrastructure can exploit to enforce correct agent behaviours and detect wrong ones. This allows us to conceptually handle a number of crucial aspects related to agent interaction, including interaction protocols, timing properties — such as delays and timeouts —, and a notion of contract featuring violation and guarantees.
Formally, our framework is based on a process algebra featuring
explicit connection between action and its related perceptions,
a time-based branch operator,
violation and guarantee deadlock states, and
association of mental properties to interactions.