Modelling Reactive Behaviour in Vertically Layered Agent Architectures

The use of layered architectures for modeling autonomous agents has become popular over the past few years. In this paper, different approaches how these architectures can be build are discussed. A special case, namely vertically layered architectures is discussed by the example of the InteRRaP agent model. The paper focusses on the lower levels of the architecture which provide reactivity, incorporate procedural knowledge, and which connect the cooperation and planning layers with the outside world. We claim that the lower system layers are likely to become a control bottleneck in vertically layered architectures, and that very careful modeling is required to produce the desired agent behaviour.