Page 301 - Introduction to AI Robotics
P. 301

284
                                                                        The Hybrid Deliberative/Reactive Paradigm
                                                                      7
                                     come by the large number of robots. The other approach, characterized by
                                     Carnegie Mellon University’s Ambler robot, proposed a single large robot
                                     with a higher viewpoint and stability. Ambler was built by “Red” Whitaker
                                     at the CMU Field Robotics Institute to be able to maintain a sensor platform
                                     at a level height by stepping over the majority of rocks, but at a tremendous
                                     penalty in size, weight, and power. In the end, planetary rover researchers
                                     have gravitated towards wheeled vehicles with some type of articulation to
                                     maintain stability, such as seen with Sandia National Laboratories’ Rattler.
                                     An extension of the Ambler design philosophy was manifested in the Dante
                                     robots. These were built to rappel down steep canyons and volcanoes on
                                     Mars (and Earth). Dante was able to lower itself successfully most of the way
                                     into a volcano in Antarctica, but could not climb back out. It was dropped
                                     while being lifted out by a helicopter, twisting its frame.


                               7.8   Evaluation of Hybrid Architectures

                                     In some regards, it is difficult to evaluate Hybrid architectures individually.
                                     Each architecture is still evolving and the deliberative component is being
                                     expanded practically daily. Returning to the four criteria set forth in the
                                     overview in Part I, it is interesting to evaluate the architectures as a whole.
                                       In terms of support for modularity, each architecture is highly modular.
                                     Most are divided into layers, which are then subdivided into modules. As
                                     the software agent programming style for AI gains in popularity, probably
                                     more architectures will implement deliberative modules as independent spe-
                                     cialists. AuRA and SFX clearly exhibit an organization which lends itself to
                                     object-oriented programming. The use of specialists also enhances the ease
                                     of portability.
                                       Hybrids tend to have a high degree of niche targetability. The addition of
                                     the deliberative component allows Hybrids to be used for applications not
                                     appropriate for purely Reactive systems. However, the partitioning between
                                     reaction and deliberation allows the reactive portion to be used alone for
                                     purely reactive applications.
                                       Another attractive aspect of Hybrid architectures is that they often explic-
                                     itly attempt to ensure robustness. In the SFX and 3T architecture, modules
                                     within the various deliberative components attempt to monitor the perfor-
                                     mance of the reactive behaviors and either replace or adapt the configuration
                                     as needed.
   296   297   298   299   300   301   302   303   304   305   306