Page 71 - Introduction to AI Robotics
P. 71

54
                                                                                  2 The Hierarchical Paradigm
                                     ing the 1970’s and 1980’s worked on either computer vision related issues,
                                     trying to get the robots to be able to better sense the world, or on path plan-
                                     ning, computing the most efficient route around obstacles, etc. to a goal lo-
                                     cation.


                               2.4   Representative Architectures

                                     As mentioned in Part I an architecture is a method of implementing a para-
                                     digm, of embodying the principles in some concrete way. Ideally, an archi-
                                     tecture is generic; like a good object-oriented program design, it should have
                                     many reusable pieces for other robot platforms and tasks.
                                       Possibly the two best known architectures of the Hierarchical period are
                                     the Nested Hierarchical Controller (NHC) developed by Meystel 93  and the
                                                                                                   1
                                     NIST Realtime Control System (RCS) originally developed by Albus, with
                                     its teleoperation version for JPL called NASREM.


                              2.4.1  Nested Hierarchical Controller
                                     As shown in Fig. 2.5, the Nested Hierarchical Controller architecture has
                                     components that are easily identified as either SENSE, PLAN, or ACT.The
                                     robot begins by gathering observations from its sensors and combining those
                                     observations to form the World Model data structure through the SENSE
                                     activity. The World Model may also contain a priori knowledge about the
                                     world, for example, maps of a building, rules saying to stay away from the
                                     foyer during the start and finish of business hours, etc. After the World
                                     Model has been created or updated, then the robot can PLAN what actions
                                     it should take. Planning for navigation has a local procedure consisting of
                                     three steps executed by the Mission Planner, Navigator, and Pilot. Each of
                                     these modules has access to the World Model in order to compute their por-
                                     tion of planning. The last step in planning is for the Pilot module to generate
                                     specific actions for the robot to do (e.g., Turn left, turn right, move straight at
                                     a velocity of 0.6 meters per second). These actions are translated into actua-
                                     tor control signals (e.g., Velocity profile for a smooth turn) by the Low-Level
                                     Controller. Together, the Low-Level Controller and actuators form the ACT
                                     portion of the architecture.
                                       The major contribution of NHC was its clever decomposition of planning
                                     into 3 different functions or subsystems aimed at supporting navigation: the
                    MISSION PLANNER  Mission Planner,the Navigator,and the Pilot. As shown in Fig. 2.6, the Mis-
                         NAVIGATOR   sion Planner either receives a mission from a human or generates a mission
                              PILOT
   66   67   68   69   70   71   72   73   74   75   76