INTELLIGENT CONTROL OF MATERIAL HANDLING SYSTEMS - Environmentally Conscious Materials Handling - page 74

Layer 3

Layer 2

Layer 1

Sensor inputs Actions

Layer 0

Figure 3.4 Brook's subsumption architecture

to simulation flocking behavior of birds and fish use a similar approach to econ-
omize with the energy consumption of the boids (Reynolds 1987). Figure 3.4
shows a model of the layered subsumption architecture by Brooks.
For cognitive and deliberative agents, the most well-known architecture that
comprises most the concepts and ideas of highly reasoning and cognitive agents
is the BDI architecture (Rao and Georgeff 1992). Here, B is for beliefs, D for
desires, and I for intensions, which very well reflect the principles of human
reasoning and other intelligent creatures. The agent has a current view of the
world or the environment in which it is situated, which is modeled through its
beliefs. The goals it has been designed to achieve are described by the desires
of the agent. One can think of the desires as a plan library or a set of described
goals that the agent wants to achieve, but not necessarily is working on at the
moment. So the decision making of the agent works by selecting the desires
that seem most achievable under the current conditions (the beliefs). When an
agent commits to pursuing a certain desire, the desire becomes an intension of
the agent, and the agent persists in pursuing this goal until it no longer appears
achievable. Thus, the agent will not just give up on a current plan, whenever
new inputs are perceived, so that challenge of using the BDI architecture is to
balance the proactive and goal-directed behavior against the influence from new
inputs, which is a more reactive behavior. A model of the BDI architecture is
presented in Figure 3.5.
Most other architectures for cognitive agents are either an extension of the
BDI architecture or they use a somewhat similar approach with a formalized
reasoning model between a set of described goals and the current world model.
Hybrid architectures, which try to combine the best of both worlds, also exist,
and the InteRRaP architecture (Figure 3.6) by Muller is a classical example of
that (Muller 1996). The InteRRaP architecture has three layers:
1. A bottom layer has all the reactive behaviors and situation-to-action rules.
This layer is also the interface to the world.