We have previously developed a navigation system for an autonomous mobile robot. It can find the landmarks on the ceiling for navigation. However, when the robot fails to recognize a landmark, the robot can not calculate the self-position. So, the robot has to prepare for the behavior of error recovery. But an error that is not able to recover may occur. In such a case, the robot needs to relocate the self-position. In this paper, we define global matching as the behavior that is able to relocate the self-position and propose to use this module for the error recovery in HALAS (hierarchical adaptive and learning architecture system) that we have proposed as a hierarchical system.

We have previously developed a navigation system for autonomous mobile robots. It can find landmarks on the ceiling to move and work. However, it is impossible to recognize landmarks when the size of landmark is different from the size of template. In such cases, the robot cannot continue to work. Therefore we proposed variable template matching (VTM) that makes it possible to recognize landmarks of different sizes. We realize VTM by using the optimization method called the evolution strategy (ES). The robot can recognize the landmark faster, inferring landmark parameters by binary density projection.

We present a vision based navigation system which has an error recovery function for an autonomous mobile robot. A misrecognition of landmark causes the robot to mislocate itself. When this happens, the robot is unable to perform its primary objective. In order to solve this kind of problem, we have proposed a hierarchical control architecture "HALAS" (hierarchical adaptive and learning architecture system). This consists of some modules which are arranged hierarchically and each module means a single function of this robot. When one module fails in its aim, the upper module detects the error and recovers from it using other useful information from its knowledge data base. We show experimentally that the robot gets a higher reliability in autonomous mobility because the map correspondence module finds the error which has been caused by an anemo perception module (landmark recognition module) and it corrects the error utilizing its map information.

A new type of mobile pipeline maintenance robot has been designed to clarify the feasibility of lazy tongs mechanism for adaptation to various size pipelines. The robot basically consists of three modules with two joints between modules and can move straight and spirally along the surface of pipeline. The robot can change its arm length in order to move on the pipeline of various size from 25 A to 100 A, and pass over obstacles such as reducer, diffuser, flange and other plant equipment. The control of the robot is so complicated that the modified dual control mode is introduced using the coordinate transformation matrix. By calculating the position and attitude of the robot under the control mode, the manoeuvrability of passing over a reducer is shown with experimental results.

We propose a vision based navigation system for autonomous mobile robots that recognizes outlets of air conditioning system (anemo) located on the ceiling as landmarks. Because landmarks on the ceiling are not obstructed by other objects, robots can find them easier than landmarks on the floor. To detect the landmark, we applied the fuzzy template matching (FTM) method. Based on FTM, a fuzzy template (FT) is modeled based on the edge information of landmarks in the image. In addition, we use neural network (NN) to detect landmarks quickly. Inputs of NN are the grades of fuzzy memberships and the learned NN directs FT to the landmark in the image. Because FTM does not use image processing, except edge extraction, it does not make processing error and the landmark can be detected precisely. The navigation system can calculate the distance and the angle of the anemo from the map information. As a result, the navigation system can identify the present robot's position and orientation. By the proposed algorithm, precise and fast landmark detection become possible even if the landmark follows the shadow.