Navigation-oriented design for in-pipe robot in recursively divided sampling space with rapidly exploring random tree

Abstract

Gas pipelines are subject to periodic inspection and maintenance for safety and longevity. Many robotic inspection systems have been developed for in-pipe applications, but systematic geometric design methodology that is suitable for in-pipe navigation has not been well studied so far due to difficulties in predicting the capability of maneuvering through the obstacles inside of pipelines such as bend, miter, and T-branch joint. The geometric design of the robot is critical to the performance of such in-pipe robots because the actuation and the measurement are constrained by the shape and the size of the robot. In this paper, we propose a design methodology that finds the maximum value of geometric design parameters of the robot with recursive evaluation of the parameter values in the design parameter space. The role of the design space division is to reduce the search region and to increase the number of parametric samples to near optimal values. As a parameter evaluation method, we adapt Rapidly exploring random tree (RRT) because it is known to be suitable for solving narrow passage problems for high-dimensional systems. Our design method makes it possible to find an optimal parameter set without computing complex cost functions. The design result of the in-pipe robot is 8 % larger than that of a heuristic geometry-based approach in three-parameter design problem.