Reference: Vina, A.; Ash, D.; & Hayes-Roth, B. Engineering Reactive Agents for Real-Time Control. Avignon, France, 1991.
Abstract: For an intelligent agent to be successful in real-time control applications, it must provide correct responses for every possible situation under the required time constraints. This paper describes how to engineer these agents to work correctly under time pressure in dynamic environments. By working correctly we mean the achievement of at least an acceptable behavior in its environment. We address the problem assuming that the following conditions hold: (1) there is no uncertainty in the agent about the goal to achieve; (2) the agent can not reliable identify the state of the environment, but it can extend its knowledge by executing some discretionary sensory operations; (3) there is uncertainty in effects of the agents actions due to the complexity of the environment; and (4) the execution of an agent's action requires time, has an associated cost, and creates some expectations about its effects. The results can be easily transferred to cases in which the agent's uncertainty is less than in the above described general scenario. We conclude the paper reviewing different approaches to the design of real-time control systems in discrete control theory and in artificial intelligence.