Behaviour Management in Real-Time Agents

In the same way living creatures follow different behaviours when facing different stimuli, some computing systems are able to benefit from using different computations for each significant situation faced. For example, this strategy is carried out as “working modes” in the Real-Time Systems domain and as “roles” in Multi-Agent Systems theory. This paper presents an extension of the ARTIS agent (AA ) architecture [1] allowing the agent to explicitly manage different behaviours. The extension has three main aspects: (1) the definition of the alternative behaviours for the agent, (2) the instruction of the agent on the particular conditions that should produce a behavioural change, and (3) the actual mechanisms making the change. Furthermore, since the AA is an architecture intended for hard real-time environments, all three aspects must be consistent with the strict temporal restrictions of the agent. Overall, the behavioural extension makes the AA architecture apter to deal with more complex and changing environments, effectively improving the adaptation capabilities of the agent, while maintaining its timeliness.