Autonomous Navigation Of Mobile Robot On Non-planar Terrain

With the development of computer science and robot technology, mobile robots are now not only widely used in industrial manufacture, but also greatly involved in military, medical and scientific fields. Autonomous navigation is one of the key aspects for mobile robots operating in wider and more complex environment which usually has non-planar terrain.First, this paper introduces the mobile platform and the sensory system which is responsible for functions such as environmental modeling. The sensory system composes of one omnidirectional and one stereo vision sensor, which is named as Heterogeneous Vision System (HVS). Then the definitions of modules including environmental modeling, real-time traversability assessment, 6-DOF self-localization and path planning are given. These four modules make up the autonomous navigation system for mobile robots operating on non-planar terrain.For environmental modeling, the definition of 2.5-dimensiional grid map is first proposed for the storage of 3-dimensional environmental data. The 2.5D grid map is companied with 2D grid map forming an environmental model called as hybrid grid map which is utilized to store the information gathered from the HVS.The mobile robot has to negotiate with terrain when operating on non-planar surface. Therefore, a real-time assessment is necessary once in order to decide whether the terrain is traversable. In this paper, the classification of terrain traversability in 2D environmental model is based on imagery segmentation according to features like color and texture. For 3D environmental model, terrain characteristics are first extracted from terrain profile and evaluated by a fuzzy logic-based classifier for traversability assessment.The state-of-the-art 6-DOF localization through omnidirectional vision requires the sensor either is equiangular or fits single viewpoint model. In this paper, a method of 6-DOF is presented through the HVS combined with a two-axis inclinometer. This method has no constraint on the equiangular property or structure of the omnidirectional vision sensor.Finally, with the results of environmental modeling and traversability assessment, the approach of path planning is discussed. Path planning aims at providing a both feasible and efficient path in the non-planar environment with no a-priori knowledge.This paper presents the solution to autonomous navigation for mobile robot working on non-planar surface. The key aspects of the navigation system: environmental modeling, traversability assessment, 6-DOF self-localization and path planning are developed step by step, through which the mobile robot is able to operate in more complex environment.