Page 306 - Introduction to Autonomous Mobile Robots

Planning and Navigation

6.3 Navigation Architectures 291

Given techniques for path planning, obstacle avoidance, localization, and perceptual inter-
pretation, how do we combine all of these into one complete robot system for a real-world
application? One way to proceed would be to custom-design an application-specific, mono-
lithic software system that implements everything for a specific purpose. This may be effi-
cient in the case of a trivial mobile robot application with few features and even fewer
planned demonstrations. But for any sophisticated and long-term mobile robot system, the
issue of mobility architecture should be addressed in a principled manner. The study of nav-
igation architectures is the study of principled designs for the software modules that con-
stitute a mobile robot navigation system. Using a well-designed navigation architecture has
a number of concrete advantages:

6.3.1 Modularity for code reuse and sharing
Basic software engineering principles embrace software modularity, and the same general
motivations apply equally to mobile robot applications. But modularity is of even greater
importance in mobile robotics because in the course of a single project the mobile robot
hardware or its physical environmental characteristics can change dramatically, a challenge
most traditional computers do not face. For example, one may introduce a Sick laser
rangefinder to a robot that previously used only ultrasonic rangefinders. Or one may test an
existing navigator robot in a new environment where there are obstacles that its sensors
cannot detect, thereby demanding a new path-planning representation.
We would like to change part of the robot’s competence without causing a string of side
effects that force us to revisit the functioning of other robot competences. For instance we
would like to retain the obstacle avoidance module intact, even as the particular ranging
sensor suite changes. In a more extreme example, it would be ideal if the nonholonomic
obstacle avoidance module could remain untouched even when the robot’s kinematic struc-
ture changes from a tricycle chassis to a differential-drive chassis.

6.3.2 Control localization
Localization of robot control is an even more critical issue in mobile robot navigation. The
basic reason is that a robot architecture includes multiple types of control functionality
(e.g., obstacle avoidance, path planning, path execution, etc.). By localizing each function-
ality to a specific unit in the architecture, we enable individual testing as well as a princi-
pled strategy for control composition. For example, consider collision avoidance. For
stability in the face of changing robot software, as well as for focused verification that the
obstacle avoidance system is correctly implemented, it is valuable to localize all software
related to the robot’s obstacle avoidance process. At the other extreme, high-level planning
and task-based decision-making are required for robots to perform useful roles in their