Page 123 - Introduction to AI Robotics
P. 123

106

                                                            Combine                  4 The Reactive Paradigm
                                                   Extract  Features  Plan    Task    Motor
                                          Sensors  Features  into     Tasks  Execution  Control  Actuators
                                                             Model



                                             SENSE                  PLAN                  ACT




                                     Figure 4.1 Horizontal decomposition of tasks into the S,P,A organization of the Hi-
                                     erarchical Paradigm.




                                       The Reactive Paradigm grew out of dissatisfaction with the hierarchical
                                     paradigm and with an influx of ideas from ethology. Although various reac-
                                     tive systems may or may not strictly adhere to principles of biological intelli-
                                     gence, they generally mimic some aspect of biology. The dissatisfaction with
                                     the Hierarchical Paradigm was best summarized by Rodney Brooks, 27  who
                         HORIZONTAL  characterized those systems as having a horizontal decomposition as shown in
                      DECOMPOSITION  Fig. 4.1.
                                       Instead, an examination of the ethological literature suggests that intelli-
                           VERTICAL  gence is layered in a vertical decomposition, shown in Fig. 4.2. Under a ver-
                      DECOMPOSITION  tical decomposition, an agent starts with primitive survival behaviors and
                                     evolves new layers of behaviors which either reuse the lower, older behav-
                                     iors, inhibit the older behaviors, or create parallel tracks of more advanced
                                     behaviors. The parallel tracks can be thought of layers, stacked vertically.
                                     Each layer has access to sensors and actuators independently of any other
                                     layers. If anything happens to an advanced behavior, the lower level be-
                                     haviors would still operate. This return to a lower level mimics degradation
                                     of autonomous functions in the brain. Functions in the brain stem (such as
                                     breathing) continue independently of higher order functions (such as count-
                                     ing, face recognition, task planning), allowing a person who has brain dam-
                                     age from a car wreck to still breathe, etc.
                                       Work by Arkin, Brooks, and Payton focused on defining behaviors and on
                                     mechanisms for correctly handling situations when multiple behaviors are
                                     active simultaneously. Brooks took an approach now known as subsumption
                                     and built insect-like robots with behaviors captured in hardware circuitry.
   118   119   120   121   122   123   124   125   126   127   128