Indoor robot navigation with a generic map

摘要

We have investigated indoor robot navigation in situations where only a generic map is available. Although some outdoor navigation methods can be used in indoor navigation, most methods for outdoor navigation are not generally applicable to indoor navigation because indoor navigation requires more precise and complete results. Indoor navigation may be performed with a specific map. The specific map, however, has several limitations. Making a specific map is difficult and tedious. In addition, it needs to be updated constantly because objects and their arrangement in an indoor setting are constantly changing. To overcome these difficulties, navigation with a generic map is proposed. The generic map includes general information about the environment. Such a generic map allows the robot to navigate in environments without knowledge of the specific instances or their locations. We have implemented a vision based robot, Antigone, on a Denning mobile robot. The robot has navigated indoor environments with only a generic map when a goal is given symbolically. Several methods for the robot have been developed concerning robustness and efficiency, which are some of the most important issues in robot navigation. (1) We have developed the Junction-supported trinocular matching method that adds robustness to a trinocular matching method. (2) We have introduced the s-map for environmental representation. The s-map can be efficiently constructed from 3-D segments. Moreover, it represents the environment more accurately in a 2-D domain consisting of width and depth, which transforms many 3-D problems in navigation into 2-D problems. (3) We have proposed an efficient generic recognition system for landmark recognition, which works in real indoor environments. (4) We have suggested the GPS-like path planning for vision based navigation, which is an efficient path planning algorithm.