On the Mapping Problem in SLAM Approaches for Autonomous Robot Navigation

Abstract Simultaneous Localization and Mapping (SLAM) is a well-known strategy for enabling robots to maneuver in unknown environments. The solution to a SLAM problem provides information about the location of the robot and the structure of the environment in a way that enables better decisions to be made by that robot. There are many ways of localizing the robot by using on-board sensors such as ranging sensors, motion sensors, and vision sensors; however, all SLAM methods tend to share features of the mapping strategy. Generally, there are two main mapping methods. One approach augments the coordinates of all landmarks in the environment with the robot pose as one state vector, and the second creates a 2-D or 3-D grid of the environment and then assigns occupancy levels of the landmarks in each grid location. Although both methods have been effective, they have some limitations. The first method is computationally burdensome since increases the size of the state vector whenever landmarks are observed; the second method treats all grid occupancy levels in binary form where all occupied grids are assumed to be at equal levels although certain grids may have occupancies that do not affect the robot mobility. This paper presents a simple but effective grid-based mapping strategy that integrates the path planning step for autonomous ground robotic vehicles. The map is presented as a 2-D surface on which each point is weighted based on observed landmarks and their sizes. This map is updated continuously as new landmarks are detected without changing memory requirements of the robot. The path planning task is implemented by joining points that are on a steep path towards the destination. Preliminary numerical results on its performance on path planning indicate the robot was able to follow the shortest path and the process is not computationally intensive.